I wonder at the end of this if it's the still worth the risk?
A lot of how I form my thoughts is driven by writing code, and seeing it on screen, running into its limitations.
Maybe it's the kind of work I'm doing, or maybe I just suck, but the code to me is a forcing mechanism into ironing out the details, and I don't get that when I'm writing a specification.
I second this. This* is the matter against which we form understanding. This here is the work at hand, our own notes, discussions we have with people, the silent walk where our brain kinda process errors and ideas .. it's always been like this since i was a kid, playing with construction toys. I never ever wanted somebody to play while I wait to evaluate if it fits my desires. Desires that often come from playing.
Outsourcing this to an LLM is similar to an airplane stall .. I just dip mentally. The stress goes away too, since I assume the LLM will get rid of the "problem" but I have no more incentives to think, create, solve anything.
Still blows my mind how different people approach some fields. I see people at work who are drooling about being able to have code made for them .. but I'm not in that group.
I'll push it back against this a little bit. I find any type of deliberative thinking to be a forcing function. I've recently been experimenting with writing very detailed specifications and prompts for an LLM to process. I find that as I go through the details, thoughts will occur to me. Things I hadn't thought about in the design will come to me. This is very much the same phenomenon when I was writing the code by hand. I don't think this is a binary either or. There are many ways to have a forcing function.
I think it's analogous to writing and refining an outline for a paper. If you keep going, you eventually end up at an outline where you can concatenate what are basically sentences together to form paragraphs. This is sort of where you are now, if you spec well you'll get decent results.
I agree, I felt this a bit. The LLM can be a modeling peer in a way. But the phase where it goes to validate / implement is also key to my brain. I need to feel the details.
My think/create/solve focus is on making my agentic coding environment produce high quality code with the least cost. Seems like a technical challenge worth playing with.
It probably helps that I have 40 years of experience with producing code the old ways, including using punch cards in middle school and learning basic on a computer with no persistent storage when I was ten.
I think I've done enough time in the trenches and deserve to play with coding agents without shame.
Alternatively, another second order effect is can't sip latte anymore because you're orchestrating 8 bots do the work and you're back to 80%-100% time saturation.
So far in my career I have always had more requests coming in than implementations going out. If I can go 3 or 10 times faster, than I will still have plenty of work. Especially for the slew of ideas that are never even considered to put towards a dev, because it's already considered to be too low value to have it even be considered to be build. Or the ideas that are so far fetched they were never considered feasible. I am not worried work will dry up.
What I believe is going to be interesting is what happens when non-engineers adopt building with agentic AI. Maybe 70 or 80% of their needs will be met without anyone else directly involved. My suspicion is that it will just create more work: making those generated apps work in a trustworthy manner, giving the agents more access to build context and make decisions, turning those one off generated apps into something maintainable, etc.
Exactly this. Even if right now you, bottom level wage earning grunt, get to lighten your workload for a fleeting second, sit back and enjoy the latte it's only but a fleeting second until the capital class tighten the screws.
Most people will get laid off and made redundant and those who remain are going to have to run faster than ever to produce wealth for the capital owners.
Yea, I don’t think that will be the case. Spreadsheets simplified the work of junior finance people who did all the work by hand before. But more people work in finance now than before.
There were always many mediocre engineers around, some of them even with fancy titles like "Senior," "Principal", and CTO.
We have always survived it, so probably we can also survive mediocre coders not reading the code the LLM generates for them because they are unable to see the problems that they were never able to see in their handwritten code.
Honestly it’s not that hard. I already coded less and less as part of my job as I get more senior and just didn’t have time, but I was still easy to do code reviews and fix bugs, sit down and whip out a thousand lines in a power session. Once you learn it doesn’t take much practice to maintain it. A lot of traditional coding is very inefficient. With AI it’s like we’re moving from combustion cars to EVs, the energy efficiency is night and day, for doing the same thing.
That said, the next generation may struggle, but they’ll find their way.
It’s going to be extremely difficult if PR and code reviews do not prune unnecessary functions. From what I’m experiencing now, there’s a lot of additional code that gets generated.
In this case why can’t other agents just automate your job completely ? They are capable of that. What do you bring in the process of still doing manual organization ?
I still have to tell it what to do, and often how to do it. I manage its external memory and guidelines, and review implementation plans. I’m still heavily involved in software design and test coverage.
AI is not capable yet of automating my job completely – I anticipate this will happen within two years, maybe even this year (I’m an ML researcher).
No, I mean that my job in its current form – as an ML researcher with a phd and 15 years of experience - will be completely automated within two years.
Is the progress of LLMs moving up abstraction layers inevitable as they gather more data from each layer? First, we fed LLMs raw text and code and now they are gathering our interactions with the LLM regarding generated code. It seems like you could then use the interactions to make a LLM that is good at prompting and fixing another LLMs generated code. Then its on to the next abstraction layer.
What you described makes sense, and it's just one of the things to try. There are lots of other research directions: online learning, more efficient learning, better loss/reward functions, better world models from training on Youtube/VR simulations/robots acting in real world, better imitation learning, curriculum learning, etc. There will undoubtedly be architectural improvements, hardware improvements, longer context windows, insights from neuroscience, etc. There is still so much to research. And there are more AI researchers now than ever. Plus current AI models already make us (AI researchers) so much more productive. But even if absolutely no further progress is made in AI research, and foundational model development stops today, there's so much improvement to be made in the tooling around the models: agentic frameworks, external memory management, better online search, better user interactions, etc. The whole LLM field is barely 5 years old.
So your assumption is that it will ultimately be the users of software themselves who will throw some every day language at an AI and it will reliably generate something that meets those users' intuitive expectations?
Yes, it will be at least as reliable as an average software engineer at an average company (probably more reliable than that), or at least as reliable as a self-driving car where a user says get me to this address, and the car does it better (statistically) than an average human driver.
I think this could work for some tasks but not for others.
We didn't invent formal languages to give commands to computers. We invented them as a tool for thinking and communicating things that are hard to express in natural language.
I doubt that we will stop thinking and I doubt that it will ever be efficient to specify tasks purely in terms of natural language.
One of my first jobs as a software engineer was for a bank (~30 years ago). This bank manager wasn't a man of many words. He just handed us an Excel sheet as a specification for what he wanted us to implement.
My job right now is to translate natural English statements from my bosses/colleagues into natural English instructions for Claude. Yes, it takes skill and experience to do this effectively. But I don't see any reasons Gemini 4, Opus 5 or GPT-6 won't be able to do this just as well as I do.
I have enough savings for a few years, so I might just move to a lower COL area, and wait it out. Hopefully after the initial chaos period things will improve.
For someone at your position with your experience it’s quite depressing that your job is going to be automated. I feel quite anxious when I see young generations in my country that say themselves they are lazy about learning new things. The next generation will be useless to capitalist societies, in a sense that they won’t be able to bring value through administrative or white collar work. I hope some areas of the industry will move slowly toward AI
Actually for me it was the opposite: before I wasn't able to play around and experiment in my free time that much, because I didn't have enough energy left to actualize the thoughts and ideas I have since I have a day job.
Now, since the bottleneck of moving the fingers to write code has gone down, I actually started to enjoy doing side projects. The mental stress from writing code has gone down drastically with Claude Code, and I feel the urge to create more nowadays!
you have a point.. i'm still confused about how this will affect jobs, markets
in a way a personal project is different from a job duty, here you're exploring, less if no deadline.. at work if I feel the llm is doing everything and I don't really master, i risk my job and my skills rot.
Everything you have said here is completely true, except for "not in that group": the cost-benefit analysis clearly favors letting these tools rip, even despite the drawbacks.
But it's also likely that these tools will produce mountains of unmaintainable code and people will get buried by the technical debt. It kind of strikes me as similar to the hubris of calling the Titanic "unsinkable." It's an untested claim with potentially disastrous consequences.
> But it's also likely that these tools will produce mountains of unmaintainable code and people will get buried by the technical debt.
It's not just likely, but it's guaranteed to happen if you're not keeping an eye on it. So much so, that it's really reinforced my existing prejudice towards typed and compiled languages to reduce some of the checking you need to do.
Using an agent with a dynamic language feels very YOLO to me. I guess you can somewhat compensate with reams of tests though. (which begs the question, is the dynamic language still saving you time?)
You can (and probably should) still do tests, but there's an entire class of errors you know can't happen, so you need far less tests, focusing only on business logic for the most part.
Static type checking is even faster than running the code. It doesn't catch everything, but if finding a type error in a fast test is good, then finding it before running any tests seems like it would be even better.
I can provide evidence for your claim. The technical debt can easily snowball if the review process is not stringent enough to keep out unnecessary functions.
Oh I'm well aware of this. I admitted defeat in a way.. I can't compete. I'm just at loss, and unless LLM stall and break for some reason (ai bubble, enshittification..) I don't see a future for me in "software" in a few years.
Somehow I appreciate this type of attitude more than the one which reflects total denial of the current trajectory. Fervent denial and AI trash-talking being maybe the single most dominant sentiment on HN over the last year, by all means interspersed with a fair amount of amazement at our new toys.
But it is sad if good programmers should loose sight of the opportunities the future will bring (future as in the next few decades). If anything, software expertise is likely to be one of the most sought-after skills - only a slightly different kind of skill than churning out LOCs on a keyboard faster than the next person: People who can harness the LLMs, design prompts at the right abstraction level, verify the code produced, understand when someone has injected malware, etc. These skills will be extremely valuable in the short to medium term AFAICS.
But ultimately we will obviously become obsolete if nothing (really) catastrophic happens, but when that happens then likely all human labor will be obsolete too, and society will need to be organized differently than exchanging labor for money for means of sustenance.
I get crazy over the 'engineer are not paid to write loc', nobody is sad because they don't have to type anymore. My two issues are it levels the delivery game, for the average web app, anybody can now output something acceptable, and then it doesn't help me conceptualize solution better, so I revert to letting it produce stuff that is not maleable enough.
I wonder about who "anybody can now output something acceptable" will hit most - engineers or software entrepreneurs.
Any implementation moat around rapid prototyping, and any fundraising moat around hiring a team of 10 to knock out your first few versions, seems gone now. Trying to sell MVP-tier software is real hard when a bunch of your potential customers will just think "thanks for the idea, I'll just make my own."
The crunch for engineers, on the other hand, seems like that even if engineers are needed to "orchestrate the agents" and manage everything, there could be a feature-velocity barrier for the software that you can still sell (either internally or externally). Changing stuff more rapidly can quickly hit a point of limited ROI if users can't adjust, or are slowed by constant tooling/workflow churn. So at some point (for the first time in many engineers' career, probably) you'll probably see product say "ok even though we built everything we want to test, we can't roll it all out at once!". But maybe what is learned from starting to roll those things out will necessitate more changes continually that will need some level of staffing still. Or maybe cheaper code just means ever-more-specialized workflows instead of pushing users to one-size-fits-all tooling.
In both of those cases the biggest challenge seems to be "how do you keep it from toppling down over time" which has been the biggest unsolved problem in consumer software development for decades. There's a prominent crowd right now saying "the agents will just manage it by continuing to hack on everything new until all the old stuff is stable too" but I'm not sure that's entirely realistic. Maybe the valuable engineering skills will be putting in the right guardrails to make sure that behavioral verification of the code is a tractable problem. Or maybe the agents will do that too. But right now, like you say, I haven't found particularly good results in conceptualizing better solutions from the current tools.
> your potential customers will just think "thanks for the idea, I'll just make my own."
yeah, and i'm surprised nobody talks about this much. prompting is not that hard, and some non software people are smart enough to absorb the necessary details (especially since the llm can tutor them on the way) and then let the loop produce the MVP.
> Or maybe cheaper code just means ever-more-specialized workflows instead of pushing users to one-size-fits-all tooling.
If the world comes to that it will be absolutely catastrophic, and it’s a failure of grappling with the implications that many of the executives of AI companies think you can paper over the social upheaval with some UBI. There will be no controlling what happens, and you don’t even need to believe in some malicious autonomous AI to see that.
The future is either a language model trained on AI code bloats and the ways to optimize the bloat away
OR,
something like Mercor, currently getting paid really well by Meta, OpenAI, Anthropic and Gemini to pay very smart humans really well to proof language model outputs.
i'm sorry if I pulled everybody down .. but it's been many months since gemini and claude became solid tools, and regularly i have this strong gut feeling. i tried reevaluating my perception of my work, goals, value .. but i keep going back to nope.
After a multi-decade career that spanned what is rapidly seeming like the golden age of software development, I have two emotions: first gratefulness; second a mixture of resignation, maudlin reflection, and bitterness that I am fighting hard to resist.
As someone who’s always wanted to “get home and code something on my own”, I do have a glimmer of hope that I wonder if others share. I’ve worked extensively with Claude and there’s no question I am now a high velocity “builder” and my broad experience has some value here. I am sad that I won’t be able to deeply look at all the code I am producing, but I am making sure the LLM and I structure things so that I could eventually dig in to modules if needed (unlikely to happen I suppose).
Anyway, my hope/question: if I embrace my new role as fast system builder and I am creative in producing systems that solve real problems “first”, is there a path to making that a career (I.e. 4 friends and I cranking out real production software that’s filling a real niche)? There must be some way for this to succeed —- I am not yet buying the “everything will be instantly copyable and so any solution is instantly commodity” argument. If that’s true, then there is no hope. I am still in shape, though, so going pro in pickleball is always an option, ha ha.
Unfortunately you aren't a high velocity builder. The velocity curve has now shifted and everyone having Claude blast out loc after loc is now a high velocity builder. And when everyone is a high velocity builder...nobody is.
Fair point, but my hope is that the creativity involved in deciding what to build, with the choice informed by engineering experience (the project/value will not be obvious to everyone) will allow differentiation.
"creativity involved in deciding what to build, with the choice informed by engineering experience (the project/value will not be obvious to everyone) will allow differentiation."
How? Anyone upon seeing your digital product can just prompt the same thing in no time. If you can prompt it, I can prompt it and so can a million other people.
Nobody whether an individual or business holds any uniqueness or advantage to themselves. All careers and skill sets are leveled and worthless. Implementation skills are worthless. Creativity is worthless.
Agree on data value, but as mentioned above I am not yet buying the “everything will be instantly copyable and so any solution is instantly commodity” argument … crud web-app sure, something with significant back-end complexity or a multi-service systems level solution, not so much. Perhaps optimistic, admittedly. Cheers.
I hear you. And maybe you're right. Maybe I'm deluding myself, but: when I look at my skilled colleagues who vibecode, I can't understand how this is sustainable. They're smart people, but they've clearly turned off. They can't answer non-trivial questions about the details of the stuff they (vibe-)delivered without asking the LLM that wrote it. Whoever uses the code downstream aren't gonna stand (or pay!) for this long-term! And the skills of the (vibe-)authors will rapidly disappear.
Maybe I'm just as naive as those who said that photographs lack the soul of paintings. But I'm not 100% convinced we're done for yet, if what you're actually selling is thinking, reasoning and understanding.
The difference with a purely still photograph is that code is a functional encoding of an intention. Code of an LLM could be perfect and still not encode the perfect intention of the product. I’ve seen that in many occasions.
Many people don’t understand what code really is about and think they have a printer toy now and we don’t have to use pencils.
That’s not at all the same thing.
Code is intention, logic, specific use case all at once. With a non deterministic system and vague prompting there will be misinterpreted intentions from LLM because the model makes decisions to move forward. The problem is the scale of it, we’re not talking about 1000 loc. In a month you can generate millions of loc, in a year hundreds of millions of loc.
Some will have to crash and burn their company before they realize that no human at all in the loop is a non sense. Let them touch fire and make up their mind I guess.
> Code is intention, logic, specific use case all at once. With a non deterministic system and vague prompting there will be misinterpreted intentions from LLM because the model makes decisions to move forward. The problem is the scale of it, we’re not talking about 1000 loc. In a month you can generate millions of loc, in a year hundreds of millions of loc.
People are also non deterministic. When I delegate work to team of five or six mid level developers or God forbid outsourced developers, I’m going to have to check and review their work too.
It’s been over a decade that my vision/responsibility could be carried out by just my own two hands and be done on time within 40 hours a week - until LLMs
Ofc people are non deterministic. But usually we expect machines to be. That’s why we trust them blindly and don’t check the calculations. We review people’s work all the time though.
Here people will stop review machine LLM code as it’s kind of a source of truth like in other areas. That’s my point, reviewing code takes time and even more time when no human wrote it. It’s a dangerous path to stop reviews because of trust in the machine now that the machine is just kind of like humans, non deterministic.
No one who has any knowledge or who has ever used an LLM expects determinism.
And there are no computer professionals who haven’t heard about hallucinations.
Reviewing whether the code meets requirements through manual and automated tests - and that’s all I cared about when I had a team of 8 under me - is the same regardless. I wasn’t checking whether John used a for loop or while loop in between my customer meetings and meetings with the CTO. I definitely wasn’t checking the SOQL (not a typo) of the Salesforce consultants we hired. I was testing inputs and outputs and UX.
Having a team of 8 people producing code is manageable. Having an AI with 8 agents that write code all day long is not the same volume it can generate more code in a day that one person can review in a week.
What you say is that, product teams will prompt what they want to a framework, the framework will take care of spec analysis, development, reviews, compliance with spec. Product teams with QA will make sure the delivery is functionally correct.
No humans need to make sure of anything code related.
What we don’t know yet is, does AI will still produce solid code trough the years because it’s all statistical analysis and with the volume of millions of loc, refactoring needed, data migrations etc what will happen ?
For context, I just started using coding agents - codex CLI and Claude code in October. Once I saw that you had to be billed by use, I’m not using my own money for it when it’s for a company.
Two things changed - Codex CLI now lets you use it with your $20 a month subscription and I have never run into quota issues with it and my employer signed up for the enterprise vs of Claude and we each have an $800 a month allowances
My argument though is “why should I care about the code?” for the most part. If I were outsourcing a project or delegating it to a team lead, I would be asking high level architectural, security and scalability questions.
AI generated the code, AI maintains the code. I am concerned about abstractions and architecture.
You shouldn’t have to maintain or refactor “millions of lines of code”, if your code is well modularized with clean interfaces, making a change for $x7 may mean making a change for $x1…$x6. But you still should be working locally in one module at the time. You should do the same for the benefit of coders. Heck my little 5 week project has three independently deployable repos in a root folder. My root Agents file just has a summary of how all three relate via a clean interface.
In the project I am working on now, besides “does it meet the requirements”, I care about security, scalability, concurrency, user experience for the end user, user experience for the operations folks when they need to make config changes, and user experience for any developers who have to make changes long after I’m off this project. I haven’t looked at a single line of code - besides the CloudFormation templates. But I can answer any architectural question about any of it. The architecture and abstractions were designed by me and dictated to the agents
On this particular project, on the coding level, there is absolutely nothing that application code like this can do that could be insecure except hypothetically embed AWS credentials into the code. But it can’t do that either since it doesn’t have access to it [1].
In this case security posture comes from the architecture - S3 block public access, well scoped IAM roles, not running “in a VPC”. Things I am checking in the infrastructure as code and I was very specific about.
The user experience has to come from design and checking manually.
I mentioned earlier that my first stab it scaled poorly. This was caused by my design and I suspected it would beforehand. But building the first version was so fast because of AI tools, I felt no pain in going with my more architecturally complicated plan B and throwing the first version away. I wouldn’t have known that by looking at the code. The code was fine it was the underlying AWS service. I could only know that by throwing 100K documents at it instead of 1000.
I designed a concurrent locking mechanism that had a subtle flaw. Throwing the code into ChatGPT into thinking mode, it immediately found it. I might have been better off just to tell the coding agents “design a locking mechanism for $x” instead of detailing it.
Even maintainability was helped because I knew I or anyone else who touched it was probably going to be using an LLM. From the get go I threw the initial contract, the discovery sessions transcripts, the design diagrams, the review of the design diagrams, my project plan and breakdown into ChatGPT and told it to render a detailed markdown file of everything - that was the beginning of my AGENTS.md file.
I asked both Codex and Claude to log everything I was doing and my decisions into separate markdown files.
Any new developer could come into my repo, fire up Claude and it wouldn’t just know what was coded, it would have full context of the project from the initial contract through to the delivery
[1] code running on AWS never explicitly has to worry about AWS credentials , the SDKs can find the information by themselves by using the credentials of the IAM role attached to the EC2 instance, Lambda, Docker container, etc.
Even locally you should be getting temporary credentials that are assigned to environment variables that the SDK retrieved automatically.
Okay - and the person ultimately leading the team is still responsibility for it whether you are delegating to more junior developers or AI. You’re still reviewing someone else’s code based on your specs
People are indeed not deterministic. But they are accountable. In the legal sense, of course, but more importantly, in an interpersonal sense.
Perhaps outsourcing is a good analogy. But in that case I'd call it outsourcing without accountability. LLMs feel more like an infinite chain of outsourcing.
As a former tech lead and now staff consultant who leads cloud implementations + app dev, I am ultimately responsible for making sure that projects are done on time, on budget and meets requirements. My manager nor the customer would allow me to say it’s one of my team members fault that something wasn’t done correctly any more than I could say don’t blame me blame Codex.
I’ve said repeatedly over the past couple of days that if a web component was done by someone else, it might as well have been created by Claude, I haven’t done web development in a decade. If something isn’t right or I need modifications I’m going to either have to Slack the web developer or type a message to Claude.
I have this nagging feeling I’m more and more skimming text, not just what the LLMs output, but all type of texts. I’m afraid people will get too lazy to read, when the LLM is almost always right. Maybe it’s a silly thought. I hope!
People will say "oh, it's the same as when the printing press came, people were afraid we'd get lazy from not copying text by hand", or any of a myriad of other innovations that made our lives easier. I think this time it's different though, because we're talking about offloading the very essence of humanity – thinking. Sure, getting too lazy to walk after cars became widespread was detrimental to our health, but if we get too lazy to think, what are we?
there are some youtube videos about the topic, be it pupil in high school addicted to llms, or adults losing skills, and not dev only, society is starting to see strange effects
I feel the same. And I expect even a lot of the early adopters and AI enthusiasts are going to find themselves as the short end of the stick sooner than later.
I've already seen this play out. The lazies in our floor were all crazy about AI because they could finally work few and finish their tasks. Until they realized that they were visibly replaceable now. The motto in team chats is "we'll lie about the productivity gains to management, just say 10% but with lots of caretaking" now
Imagine everyone who is in less technical or skilled domains.
I can't help but resist this line of thinking as a result. If the end is nigh for us, it's nigh for everyone else too. Imagine the droves of less technical workers in the workforce who will be unseated before software engineers. I don't think it is tenable for every worker in the first world to become replaced by a computer. If an attempt at this were to occur, those smart unemployed people would be a real pain in the ass for the oligarchs.
I think you have every right to doubt those telling us that they run 5 agents to generate a new SAAS-product while they are sipping latté in a bar. To work like that I believe you'll have to let go of really digging into the code, which in my experience is needed if want good quality.
Yet I think coding agents can be quite a useful help for some of the trivial, but time consuming chores.
For instance I find them quite good at writing tests. I still have to tweak the tests and make sure that they do as they say, but overall the process is faster IMO.
They are also quite good at brute-forcing some issue with a certain configuration in a dark corner of your android manifest. Just know that they WILL find a solution even if there is none, so keep them on a leash!
Today I used Claude for bringing a project I abandoned 5 years ago up to speed. It's still at work in progress, but the task seemed insurmountable (in my limited spare time) without AI, now it feels like I'm half-way there in 2-3 hours.
I think we really need to have a serious think of what is "good quality" in the age of coding agents. A lot of the effort we put into maintaining quality has to do with maintainability, readability etc. But is it relevant if the code isn't for humans? What is good for a human is not what is good for an AI necessarily (not to say there is no overlap). I think there are clearly measurable things we can agree still apply around bugs, security etc, but I think there are also going to be some things we need to just let go of.
The implications to your statement seems to me that is: "you'll never have to directly care about it yourself, so why do you care about it?". Unless you were talking about the codebase in a user-application relationship which in this case feel free to ignore the rest of my post.
I don't believe that the code will become an implementation detail, ever. When all you do is ship an MVP to demonstrate what you're building then no one cares, before or after LLM assistance. But any codebase that lives more than a year and serves real users while generating revenue deserves to have engineers who knows what's happening beyond authoring markdown instructions to multiple agents.
Your claim seems to push us towards a territory where externalizing out thought processes to a third party is the best possible outcome for all parties, because the models will only get better and stay just as affordable.
I will respond to that by pointing out that, models that will ultimately be flawless in code generation will worth a fortune in terms of adding value, and any corporation that will win the arms race will be actually killing themselves by not raising the cost of access to their services by a metric ton. This is because there will be few LLM providers that actually worth it by then, and because oligopoly is a thing.
So no. I don't expect that we'll ever reach a point where the average person will be "speaking forth" software the same way they post on Reddit, without paying cancer treatment levels of money.
But even if it's actually affordable... Why would I ever want to use your app instead of just asking an LLM to make me one from scratch? No one seems to think about that.
i've been building agent tooling for a while and this is the question i keep coming back to. the actual failure mode isn't messy code, agents produce reasonably clean, well-typed output these days. it's that the code confidently solves a different problem than what you intended. i've had an agent refactor an auth flow that passed every test but silently dropped a token refresh check because it "simplified" the logic. clean code, good types, tests green, security hole. so for me "quality" has shifted from cyclomatic complexity and readability scores to "does the output behaviour match the specification across edge cases, including the ones i didn't enumerate." that's fundamentally an evaluation problem, not a linting problem.
This is where I think its going, it feels that in the end we will end up with an "llm" language, one that is more suited to how an llm works and less human.
You can’t drop anything as long as a programmer is expected to edit the source code directly. Good luck investigating a bug when the code is unclear semantically, or updating a piece correctly when you’re not really sure it’s the only instance.
I think that's the question. Is a programmer expected to ever touch the source code? Or will AI -- and AI alone -- update the code that it generated?
Not entirely unlike other code generation mechanisms, such as tools for generating HTML based on a graphical design. A human could edit that, but it may not have been the intent. The intent was that, if you want a change, go back to the GUI editor and regenerate the HTML.
> Not entirely unlike other code generation mechanisms, such as tools for generating HTML based on a graphical design. A human could edit that, but it may not have been the intent. The intent was that, if you want a change, go back to the GUI editor and regenerate the HTML.
We largely moved back away from "work in a graphic tool then spit out HTML from it" because it wasn't robust for the level of change/iteration pace, this wasn't exactly my domain but IIRC there were especially a lot of problems around "small-looking changes are now surprisingly big changes in the generated output that have a large blast radius in terms of the other things (like interactivity) we've added in."
Any time you do a refactor that changes contract boundaries between functions/objects/models/whatever, and you have to update the tests to reflect this, you have a big risk of your new tests not covering exactly the same set of component interactions that your old tests did. LLM's don't change this. They can iterate until the tests are green, but certain changes will require changing the tests, and now "iterating until the tests are green" could be resolved by changing the tests in a way that subtly breaks surprising user-facing things.
The value of good design in software is having boundaries aligned with future desires (obviously this is never perfect foresight) to minimize that risk. And that's the scary thing to myself about not even reading the code.
So like we went from assembler to higher level programming languages, we will now move to specifications for LLMs? Interesting thought... Maybe, once the "compilers" get good enough, but for mission critical systems they are not nearly good enough yet.
Right. I work in aerospace software, and I do not know if this option would ever be on the table. It certainly isn't now.
So I think this question needs to be asked in the context of particular projects, not as an industry-wide yes or no answer. Does your particular project still need humans involved at the code level? Even just for review? If so, then you probably ought to retain human-oriented software design and coding techniques. If not, then, whatever. Doesn't matter. Aim for whatever efficiency metric you like.
It’s also pretty close to Steve Jobs initial vision of computing in the future (https://stevejobsarchive.com/stories/objects-of-our-life, 1983) but my point is that whatever it is we call AI now became reality so much faster than anyone really saw coming. Even if the pace slows down, and it didn’t yet, things are improving so massively all the time that the world can’t keep up changing to accommodate.
> I think you have every right to doubt those telling us that they run 5 agents to generate a new SAAS-product while they are sipping latté in a bar. To work like that I believe you'll have to let go of really digging into the code, which in my experience is needed if want good quality.
Also we live in a capitalist society. The boss will soon ask: "Why the fuck am I paying you to sip a latte in a bar? While am machine does your work? Use all your time to make money for me, or you're fired."
AI just means more output will be expected of you, and they'll keep pushing you to work as hard as you can.
> AI just means more output will be expected of you, and they'll keep pushing you to work as hard as you can.
That’s a bit too cynical for me. After all, yes, your boss is not paying you for sipping lattes, but for producing value for the company. If there is a tool that maximises your output, why wouldn’t he want you to use that to great efficiency?
Put differently, would a carpenter shop accept employees rejecting the power saw in favour of a hand saw to retain their artisanal capability?
> why wouldn’t he want you to use that to great efficiency
Because I deny that? It's not fun for me.
> would a carpenter shop accept employees rejecting the power saw in favour of a hand saw to retain their artisanal capability?
Why not? If that makes enough money to keep going.
You might argue that in theoretical ideal market companies who're not utilizing every possible trick to improve productivity (including AI) will lose competition, but let's be real, a lot of companies are horribly inefficient and that does not make them bankrupt. The world of producing software is complicated.
I know that I deliver. When I'm asked to write a code, I deliver it and I responsible for it. I enjoy the process and I can support this code. I can't deliver with AI. I don't know what it'll generate. I don't know how much time would it take to iterate to the result that I precisely want. So I can't longer be responsible for my own output. Or I'd spend more time baby-sitting AI than it would take me to write the code. That's my position. Maybe I'm wrong, they'll fire me and I'll retire, who knows. AI hype is real and my boss often copy&pasting ChatGPT asking me to argue with it. That's super stupid and irritating.
I started this career because I liked writing code. I no longer write a lot of code as a lead, but I use writing code to learn, to gain a deeper understanding of the problem domain etc. I'm not the type who wants to write specs for every method and service but rather explore and discover and draft and refactor by... well, coding. I'm amazed at creating and reading beautiful, stylish, working code that tells a story.
If that's taken away, I'm not sure how I could retain my interest in this profession. Maybe I'll need to find something else, but after almost a decade this will be a hard shift.
I totally emphasise as a fellow developer, but I doubt you realise what an incredibly privileged position it is to just refuse working if you don't have fun doing it. And it doesn't really make for a convincing argument to keep you employed either.
> Why not? If that makes enough money to keep going.
If all other competing carpenters use power tools, you're going to loose contracts. We've had a few incredibly easy decades as software developers where market pressure wasn't really a thing, but that is about to change when the cost of producing code drops considerably.
> You might argue that in theoretical ideal market companies who're not utilizing every possible trick to improve productivity (including AI) will lose competition […]
You're moving the goalposts here. We're not talking about wringing every last drop of efficiency out of employees. We're talking about businesses not tolerating paying for licenses for AI agents to enable developers to sip Lattés while their computer does their job. That's a fundamentally different proposition.
> That’s a bit too cynical for me. After all, yes, your boss is not paying you for sipping lattes, but for producing value for the company. If there is a tool that maximises your output, why wouldn’t he want you to use that to great efficiency?
Sitting in a cafe enjoying a latte is not "producing value for the company." If having "5 agents to generate a new SAAS-product" matches your non-AI capacity and gives you enough free time to relax in a cafe, he's going to want to you run 50 agents generating 5 new SAAS products, until you hit your capacity.
If he doesn't need 5 new SAAS products, just one, then he's going to fire you or other members of your team.
Think of it this way: you're a piece of equipment to your boss, and every moment he lets you sit idle (on the clock) is money lost. He wants to run that piece of equipment as hard as he can, to maximize his profit.
I still do this, but when I'm reviewing what's been written and / or testing what's been built.
How I see it is we've reverted back to a heavier spec type approach, however the turn around time is so fast with agents that it still can feel very iterative simply because the cost of bailing on an approach is so minimal. I treat the spec (and tests when applicable) as the real work now. I front load as much as I can into the spec, but I also iterate constantly. I often completely bail on a feature or the overall approach to a feature as I discover (with the agent) that I'm just not happy with the gotchas that come to light.
AI agents to me are a tool. An accelerator. I think there are people who've figured out a more vibey approach that works for them, but for now at least, my approach is to review and think about everything we're producing, which forms my thoughts as we go.
>but the code to me is a forcing mechanism into ironing out the details, and I don't get that when I'm writing a specification.
This is so on point. The spec as code people try again and again. But reality always punches holes in their spec.
A spec that wasn't exercised in code, is like a drawing of a car, no matter how detailed that drawing is, you can't drive it, and it hides 90% of the complexity.
To me the value of LLMs is not so much in the code they write. They're usually to verbose, start building weird things when you don't constantly micromanage them.
But you can ask very broad questions, iteratively refine the answer, critique what you don't like. They're good as a sounding board.
I love using LLMs as well as rubber ducks - what does this piece of code do? How would you do X with Y? etc.
The problem is that this spec-driven philosophy (or hype, or mirage...) would lead to code being entirely deprecated, at least according to its proponents. They say that using LLMs as advisors is already outdated, we should be doing fully agentic coding and just nudge the LLM etc. since we're losing out on 'productivity'.
>They say that using LLMs as advisors is already outdated, we should be doing fully agentic coding and just nudge the LLM etc. since we're losing out on 'productivity'.
As long as "they" are people that either profit from FOMO or bad developers that still don't produce better software than before, I'm ok ignoring the noise.
Historically software engineering has been seen as "assembly line" work by a lot of people (see all the efforts to outsource it through spec handoffs and waterfall through the years) but been implemented in practice as design-as-you-build (nobody anticipates all the questions or edge cases in advance, software specs are often an order of magnitude simpler than the actual number of branches in the code).
For mission-critical applications I wonder if making "writing the actual code" so much cheaper means that it would make more sense to do more formal design up front instead, when you no longer have a human directly in the loop during the writing of the code to think about those nasty pops-up-on-the-fly decisions.
> software specs are often an order of magnitude simpler than the actual number of branches in the code
Love this! Be it design specs or a mock from the designer. So many unaccounted for decisions. Good devs will solve many on their own, uplevel when needed, and provide options.
And absolutely it means more design up front. And without human in the direct loop, maybe people won’t skimp on this!
I also second this. I find that I write better by hand, although I work on niche applications it’s not really standard crud or react apps. I use LLMs in the same way i used to used stack overflow, if I go much farther to automate my work than that I spend more time on cleanup compared to if I just write code myself.
Sometimes the AI does weird stuff too. I wrote a texture projection for a nonstandard geometric primitive, the projection used some math that was valid only for local regions… long story. Claude kept on wanting to rewrite the function to what it thought was correct (it was not) even when I directed to non related tasks. Super annoying. I ended up wrapping the function in comments telling it to f#=% off before it would leave it alone.
> I use LLMs in the same way i used to used stack overflow, if I go much farther to automate my work than that I spend more time on cleanup compared to if I just write code myself.
yea, same here.
i've asked an ai to plan and setup some larger non straight forwards changes/features/refactorings but it usually devolves into burning tokens and me clicking the 'allow' button and re-clarifying over and over when it keeps trying to confirm the build works etc...
when i'm stuck though, or when im curious of some solution it usually opens the way to finish the work similar to stack overflow
Using AI or writing your own code isn't an xor thing. You can still write the code but have a coding assistant or something an alt/cmd-tab away. I enjoy writing code, it relaxes me so that's what I do but when I need to look something up or i'm not clear on the syntax for some particular operation instead of tabbing to a browser and google.com I tab to the agent and ask it to take a look. For me, this is especially helpful for CSS and UI because I really suck at and dislike that part of development.
I also use these things to just plan out an approach. You can use plan mode for yourself to get an idea of the steps required and then ask the agent to write it to a file. Pull up the file and then go do it yourself.
In 1987 when I first started coding, I would either write my first attempt in BASIC and see it was too slow and rewrite parts in assembly or I would know that I had to write what I wanted from the get go in assembly because the functionality wasn’t exposed at all in BASIC (using the second 64K of memory or using double hires graphics).
This past week, I spent a couple of days modifying a web solution written by someone else + converting it from a Terraform based deployment to CloudFormation using Codex - without looking at the code as someone who hasn’t done front in development in a decade - I verified the functionality.
More relevantly but related, I spent a couple of hours thinking through an architecture - cloud + an Amazon managed service + infrastructure as code + actual coding, diagramming it, labeling it , and thinking about the breakdown and phases to get it done. I put all of the requirements - that I would have done anyway - into a markdown file and told Claude and Codex to mark off items as I tested each item and summarize what it did.
Looking at the amount of work, between modifying the web front end and the new work, it would have taken two weeks with another developer helping me before AI based coding. It took me three or four days by myself.
The real kicker though is while it worked as expected for a couple of hundred documents, it fell completely to its knees when I threw 20x documents into the system. Before LLMs, this would have made me look completely incompetent telling the customer I now wasted two weeks worth of time and 2 other resources.
Now, I just went back to the literal drawing board, rearchitected it, did all of the things with code that the managed services abstracted away with a few tweaks, created a new mark down file and was done in a day. That rework would have taken me a week by itself. I knew the theory behind what the managed service was doing. But in practice I had never done it.
It’s been over a decade where I was responsable for a delivery that I could do by myself without delegating to other people or that was simple enough that I wouldn’t start with a design document for my own benefit. Now within the past year, I can take on larger projects by myself without the coordination/“mythical man Month” overhead.
I can also in a moment of exasperation say to Codex “what you did was an over complicated stupid mess, rethink your implementation from first principles” without getting reported to HR.
There is also a lot of nice to have gold plating that I will do now knowing that it will be a lot faster
That's because many developers are used to working like this.
With AI, the correct approach is to think more like a software architect.
Learning to plan things out in your head upfront without to figure things out while coding requires a mindset shift, but is important to work effectively with the new tools.
To some this comes naturally, for others it is very hard.
I dont think any complex plan should be planned in your head. But drawing diagrams, sketching components, listing pros and cons, 100%. Not jumping directly into coding might look more like jumping into spec writing a poc
Maintaining a 'mental RAM Cache' is a powerful tool to understanding the system as a whole on a deep and intuitive level, even if you can only 'render' sections at a time. The bigger it is the more you can keep track of to be able to foresee interactions between distant pieces.
It shouldn't be your only source of a plan as you'd likely wind up dropping something, but figuring out how to jiggle things around before getting it 'on paper' is something I've found helpful.
Following the RAM analogy, this sounds like saving files only in RAM, instead of creating the files in the file system, persisted on disk, and then caching it in RAM.
Personally, for me without writing or sketching I cannot think complex things: as in complex logic, constraints, etc.
I guess this is topic too abstract, so we can read into it different things.
I think what GP is referring too are technical semantics and accidental complexity. You can’t plan for those.
The same kind of planning you’re describing can and do happen sans LLM, usually on the sofa, or in front of a whiteboard. Or by reading some research materials. No good programmer rushes to coding without a clear objective.
But the map is not the territory. A lot of questions surface during coding. LLMs will guess and the result may be correct according to the plan, but technically poor, unreliable, or downright insecure.
> A lot of how I form my thoughts is driven by writing code, and seeing it on screen, running into its limitations.
If you need that, don't use AI for it. What is it that you don't enjoy coding or think it's tangential to your thinking process? Maybe while you focus on the code have an agent build a testing pipeline, or deal with other parts of the system that is not very ergonomic or need some cleanup.
this is the right answer, but many companies mandate to use ai (burn x tokens and y percent of code) now, so people are bound to use it where it might not fit
I think of it differently: I’ve been coding so long that ironing out the details and working through the specification with AI comes extremely naturally. It’s like how I would talk to a colleague and iterate on their work.
However, the quality of the code produced by LLMs needs to be carefully managed to assure it’s of a high standard. That’s why I formalized a system of checks and balances for my genetic coding that contains architectural guidelines as well as language, specific taste advice.
> A lot of how I form my thoughts is driven by writing code, and seeing it on screen, running into its limitations.
Two principles I have held for many years which I believe are relevant both to your sentiment and this thread are reproduced below. Hopefully they help.
First:
When making software, remember that it is a snapshot of
your understanding of the problem. It states to all,
including your future-self, your approach, clarity, and
appropriateness of the solution for the problem at hand.
Choose your statements wisely.
And:
Code answers what it does, how it does it, when it is used,
and who uses it. What it cannot answer is why it exists.
Comments accomplish this. If a developer cannot be bothered
with answering why the code exists, why bother to work with
them?
To your first point - so are my many markdown files that I tell Codex/Claude to keep updated while I’m doing my work including telling them to keep them updated with why I told them to do certain things. They have detailed documentation of my initial design goals and decisions that I wrote myself.
Actually those same markdown files answer the second question.
> If a developer cannot be bothered with answering why the code exists, why bother to work with them?
Most people can't answer why they themselves exist, or justify why they are taking up resources rather than eating a bullet and relinquishing their body-matter.
According to the philosophy herein, they are therefore worthless and not worth interacting with, right?
Some people like to lay the brick, some people like to draw the blueprints. I don’t think there is anything wrong with not subscribing to this onslaught on AI tooling, doing the hard work is rewarding. Whether AI will become a standard in how code is written in the future is still to be determined and I think there is a real chance that is where it goes, it shouldn’t hinder your love for doing what you do.
100%. To me the real question is whether all the bother getting the agents to not waste time nets out to real gains, or perceived gains (while possibly even losing efficiency).
It's not at all clear to me which is true given the level of hype and antipathy out there. I'm just going to watch and wait, and experiment cautiously, till it's more clearcut.
I liken it to manual versus automated industrial production. I think manual coding will always have its place just like how there are even still people who craft things by manual labor, whether it’s woodworkers only using manual tools or blacksmiths who still manually stoke coke fires that produce very unique and custom products; vs the highly automated production lines we have that produce acceptable forms of something efficiently, and many of them so many people can have them.
This is exactly the issue I’m facing especially when working with AI-generated codebases.
Coding is significantly faster but my understanding of the system takes a lot longer because I’m having to merge my mental model with what was produced.
> A lot of how I form my thoughts is driven by writing code, and seeing it on screen, running into its limitations.
I completely agree but my thought went to how we are supposed to estimate work just like that. Or worse, planning poker where I'm supposed to estimate work someone else does.
i go back and forth on this. when i'm working on something where the hard part is the actual algorithm, say custom scheduling logic or a non-trivial state machine, i need my hands in the code because the implementation is the thinking. but for anything where the complexity is in integration rather than logic, wiring up OAuth flows, writing CRUD endpoints, setting up CI pipelines, agents save me hours and the output is usually fine after one review pass. the "code as thought" argument is real but it applies to maybe 20% of what most of us ship day to day. the other 80% is plumbing where the bottleneck is knowing what to build, not how.
I sometimes wonder if the economics of AI coding agents only work if you totally ignore all the positive externalities that come with writing code.
Is the entire AI bubble just the result of taking performance metrics like "lines of code written per day" to their logical extreme?
Software quality and productivity have always been notoriously difficult to measure. That problem never really got solved in a way that allowed non technical management to make really good decisions from the spreadsheet level of abstraction... but those are the same people driving adoption of all these AI tools.
Engineers sometimes do their jobs in spite of poor incentives, but we are eliminating that as an economic inefficiency.
I dunno. On the one hand, I keep hearing anecdata, including hackernews comments, friends, and coworkers, suggesting that AI-assisted coding is a literal game changer in terms of productivity, and if you call yourself a professional you'd better damn well lock the fuck in and learn the tools. At the extreme end this takes the form of, you're not a real engineer unless you use AI because real engineering is about using the optimal means to solve problems within time, scale, and budget constraints, and writing code by hand is now objectively suboptimal.
On the other hand, every time the matter is seriously empirically studied, it turns out that overall:
* productivity gains are very modest, if not negative
* there are considerable drawbacks, including most notably the brainrot effect
Furthermore, AI spend is NOT delivering the promised returns to the extent that we are now seeing reversals in the fortunes of AI stocks, up to and including freakin' NVIDIA, as customers cool on what's being offered.
So I'm supposed to be an empiricist about this, and yet I'm supposed to switch on the word of a "cool story bro" about how some guy built an app or added a feature the other day that he totally swears would have taken him weeks otherwise?
I'm like you. I use code as a part of my thought process for how to solve a problem. It's a notation for thought, much like mathematical or musical notation, not just an end product. "Programs must be written for people to read, and only incidentally for machines to execute." I've actually come to love documenting what I intend to do as I do it, esp. in the form of literate programming. It's like context engineering the intelligence I've got upstairs. Helps the old ADHD brain stay locked in on what needs to be done and why. Org-mode has been extremely helpful in general for collecting my scatterbrained thoughts. But when I want to experiment or prove out a new technique, I lean on working directly with code an awful lot.
I was just thinking this the other day after I did a coding screen and didn't do well. I know the script for the interviewee is your not suppsed to write any code until you talk through the whole thing, but I think i woukd have done better if I could have just wrote a bunch of throw away code to iterate on.
Are there still people under the impression that the correct way to use Stack Overflow all these years was to copy & paste without analyzing what the code did and making it fit for purpose?
If I have to say, we're just waiting for the AI concern caucus to get tired of performing for each other and justifying each other's inaction in other facets of their lives.
The post touches very briefly on linting in 7. For me, setting up a large number of static code analysis checks has had the highest impact on code quality.
My hierarchy of static analysis looks like this (hierarchy below is Typescript focused but in principle translatable to other languages):
9. Custom script to ensure shared/util directories are not over stuffed (built this using dependency-cruiser as a library rather than an exec)
10. Security check (semgrep)
I stitch all the above in a single `pnpm check` command and defined an agent rule to run this before marking task as complete.
Finally, I make sure `pnpm check` is run as part of a pre-commit hook to make sure that the agent has indeed addressed all the issues.
This makes a dramatic improvement in code quality to the point where I'm able to jump in and manually modify the code easily when the LLM slot machine gets stuck every now and then.
(Edit: added mention of pre-commit hook which I missed mention of in initial comment)
this is close to what i've landed on too. the pre-commit hook is non-negotiable. i've had Claude Code report "all checks pass" when there were 14 failing eslint rules. beyond the static analysis though, i keep hitting a harder problem: code that passes every lint rule, compiles clean, and greens the test suite but implements a subtly wrong interpretation of the spec. like an API handler that returns 200 with an empty array instead of 404, technically valid but semantically wrong. evaluating behavioural correctness against intent, not just syntax or type safety, is the gap nobody's really cracked yet. property-based testing helps but it still requires you to formalize the invariants upfront, which is often the hard part.
Not a catch all to fix issues agree with linting. Being very strict with linters has become very cheap with coding agents and it keeps you up to date with code standards and keeps code style homogenous which is very nice when you are reviewing professional code, regardless of who wrote it.
It’s also tricky otherwise if you have to occasionally review lazily written manual code mixed with syntactically formal/clean but functionally incorrect AI code.
I use a pre-commit hook to run `pnpm check`. I missed mentioning it in the original comment. Your reply reminded me of it and I have now added it. Thanks.
That's something I find to be incredibly frustrating. I have to keep reminding it that we're not done, no matter how much I enforce that the lints must pass before we're done.
These kinda things aren’t really the issues I run into. Lack of clarity of thought, overly verbose code, needlessly defensive programming - the stuff that really rots a codebase. Honestly some of the above rules you have I’d want the LLM to ignore at the times if we’re going for maximum maintainability.
The real value that AI provides is the speed at which it works, and its almost human-like ability to “get it” and reasonably handle ambiguity. Almost like tasking a fellow engineer. That’s the value.
By the time you do everything outlined here you’ve basically recreated waterfall and lost all speed advantage. Might as well write the code yourself and just use AI as first-pass peer review on the code you’ve written.
A lot of the things the writer points out also feel like safeguards against the pitfalls of older models.
I do agree with their 12th point. The smaller your task the easier to verify that the model hasn’t lost the plot. It’s better to go fast with smaller updates that can be validated, and the combination of those small updates gives you your final result. That is still agile without going full “specifications document” waterfall.
It’s a solid post overall and even for people with a lot of experience there’s some good ideas in here. “Identify and mark functions that have a high security risk, such as authentication, authorization” is one such good idea - I take more time when the code is in these areas but an explicit marking system is a great suggestion. In addition to immediate review benefits, it means that future updates will have that context.
“Break things down” is something most of us do instinctively now but it’s something I see less experienced people fail at all the time.
Brain surgery is probably a bad example... or maybe a good one, but for different reasons?
Brain surgery is highly technical AND highly vibe based.
You need both in extremely high quantities. Every brain is different, so the super detailed technical anatomies that we have is never enough, and the surgeon needs constant feedback (and insanely long/deep focus).
I can't help but keep finding it ridiculous how everyone now discovers basic best practices (linting, documentation, small incremental changes) that have been known for ages. It's not needed because of AI, you should have been doing it like this before as well.
Anyone who’s been a developer for more than 10 minutes knows that best practices are hard to always follow through on when there’s pressure to ship.
But there’s more time to do some of these other things if the actual coding time is trending toward zero.
And the importance of it can go up with AI systems because they do actually use the documentation you write as part of their context! Direct visible value can lead people to finally take more seriously things that previously felt like luxuries they didn’t have time for.
Again if you’ve been a developer for more than 10 minutes, you’ve had the discouraging experience of pain-stakingly writing very good documentation only for it to be ignored by the next guy. This isn’t how LLMs work. They read your docs.
> Anyone who’s been a developer for more than 10 minutes knows that best practices are hard to always follow through on when there’s pressure to ship.
>
But there’s more time to do some of these other things if the actual coding time is trending toward zero.
I think you'll find even less time - as "AI" drives the target time to ship toward zero.
I agree that this will be the end result over time, maybe even faster than we expect. And as those speed pressures increase, AI will take over more and more of the development process.
Religiously, routinely refactor. After almost every feature I do a feature level code analysis and refactoring, and every few features - codebase wide code analysis and refactoring.
I am quite happy with the resulting code - much less shameful than most things I've created in 40 years of being passionate about coding.
This. Historically there's been a lot of resistance to the idea of refactoring or refining features. The classic "It works, just ship it" mentality that leaves mountains of tech debt in its wake.
And there _was_ a good reason to resist refactoring. It takes time and effort! After "finishing" something, the timeline, the mental and physical energy, the institutional support, is all dried up. Just ship it and move on.
But LLMs change the equation. There's no reason to leave sloppy sub-optimal code around. If you see something, say something. Wholesale refactoring your PR is likely faster than running your test suite. Literally no excuses for bad code anymore.
You'd think it didn't need to be said but, given we have a tool to make coding vastly more efficient, some people use that tool to improve quality rather than just pump out more quantity.
We are becoming spec writers, wearing the PM/lead hats.
1) Do a gap and needs assessment.
2) Build business requirements.
3) Define scope of work to advance fulfillment.
4) Create functional and non-functional specs.
5) Divide-conquer-refine loop.
This is the main thing I have learned too. I've been building an internal tool for myself to annotate lines in each commit diff as good (green) / needs refactor (yellow) / needs rewrite (red) and it has helped me keep track of this kind of tech debt. Basically does what you could do with "TODO refactor" comments all over, but is more comprehensive and doesn't litter your source code. Plan to open source it once I've dog-fooded it a little more
Remember having to write detailed specs before coding? Then folks realized it was faster and easier to skip the specs and write the code? So now are we back to where we were?
One of the problems with writing detailed specs is it means you understand the problem, but often the problem is not understand - but you learn to understand it through coding and testing.
Skip specs, and you often ended up writing the wrong program - at substantial cost.
The main difference now is the parrots have reduced the cost of the wrong program to near zero, thereby eliminating much of the perceived value if a spec.
We’re not „thinking with portals” about these things enough yet. Typically we’d want a detailed spec beforehand, as coding is expensive and time consuming, thus we want to make sure we’re coding the right thing. With AI though, coding is cheap. So let AI skip the spec and write the code badly. Then have it review the solution, build understanding, design a spec for better solution and have it write it again. Rinse and repeat as many times you need.
It’s also nothing new, as it’s basically Joe Armstrong's programming method. It’s just not prohibitively expensive for the first time in history.
Too bad that software developers are carrying water for those who hate them and mock them for being obsolete in 6-12 months, while they are eating caviar (probably evading sanctions) and clink the champagne glasses in Davos:
Nobody wants meaningless preventable toil. What people want is a living. Nobody would be afraid of AI taking their jobs if it didn’t mean that they’d get fired.
The enthusiasm so many devs show for it is also quite bizarre, saying things like "AI makes me so much more productive," with the implication that they will be its primary beneficiaries, and that it won't result in a massive reduction in demand, compensation, and status for developers, adversely affecting them. Even more bizarre when you realize these devs aren't the ones optimizing some popular video codec or writing avionics software for a fighter jet, but instead gluing together NPM packages--probably the first or second rung on on the software "innovator's dilemma" ladder of disruption.
Spec-driven development is the only reliable way to work with AI. That's my current understanding. I spend more time refining the spec and bouncing ideas off of AI/team than before, which is good before there can't be any incorrect assumptions or hidden variables, otherwise AI will create suboptimal code. We should have been doing this much earlier in the process, even without AI, but now it's more necessary than ever. If you keep asking AI to make small changes as you learn about the business domain of your project, it will create a mess, in my experience. It's better to start from scratch and ask it to reimplement, if you finally understand all the requirements.
Sentiments like this make me wonder if perhaps the dream of the 90s was just ahead of its time. Things like UML, 4GLs, Rational were all being hyped. We were told that the future was a world where people could express the requirements & shape of the system, and the machines would do the rest.
Clearly that didn't happen, and then agile took over from the more waterfall/specs based approaches, and the rest was history.
But now we're entering a world where the state of the art is expressing your requirements & shape of the system. Perhaps this is just part of a broader pendulum swing, or perhaps the 1990s hopes & dreams finally caught up with technology.
Yes and no I'd say.
It's still the case that now only by iterating and testing things with the AI you get closer to an actually good solution.
So up front big spec will also not work so well.
The only exception maybe if you already have a very clear understanding and existing tests (like what they did with the Claude's building the rust c compiler to compile the Linux kernel)
I think PG said something about sitting down and hacking being how you understand the problem, and it’s right. You can write UML after you’ve got your head round it, but the feedback loop when hacking is essential.
I'm still writing code. I'm doing it to solve a problem, there's more to writing code than than typing. Recently AI massively simplified "getting started", and all of the tips here are applicable to working well on a team.
My recent experience: I'm porting an app to Mac. It's been in my backlog for ~2 years. With Claude I had a functional prototype in under a day getting the major behavior implemented. I spent the next two weeks refactoring the original app to share as much logic as possible. The first two days was lots of fun. The refactoring was also something I wanted to flush out unit tests, still enjoyable.
The worst part was debugging really bugs introduced to my code from 5 years ago. My functions had naming issues describing the behavior wrong, confusing Claude, that I needed to re-understand to add new features.
Parts of coding are frustrating. Using AI is frustrating for different reasons.
The most frustrating part was rebasing with git to create a sensible history (which I've had to do without AI in the past), reviewing the sheer volume of changes (14k lines) and then deciding "do I want my name on this" which involved cleaning up all the linter warnings I'd self imposed on myself.
I’m finding it to be the opposite. I used to love writing everything by hand but now Claude is giving me the ability to focus more on architecture. I like just sitting down with my coffee and thinking about the next part of my project, how I’d like it to be written and Claude just fills it in for me. It makes mistakes at times but it also finds a lot of mine that I hadn’t even realized were in my code base.
Yep, I get that some people love the act of literally typing "x = 2;" but to me coding is first and foremost problem solving. I have a problem (either truly mine or someone else's), I come up with a solution in my head and slowly implement it.
Before I also had to code it and then make sure it had no issues.
Now I can skip the coding and then just have something spit out something which I can evaluate whether I believe is a good implementation of my solution or not.
Of course, you need the skill to know good from bad but for medium to senior devs, AI is incredibly useful to get rid of the mundane task of actually writing code, while focusing on problem solving with critical review of magically generated code.
A good bit of scaffolding and babysitting allows you to let the model run much faster and more efficiently. Building your tool faster. I don't code to code, I code to build something I want.
Also there is no "compiler" and "type checker" for your SPEC. If you get something wrong in some paragraph somewhere and or contradict something in your spec X paragraphs later - you have to use Mark-1 EyeBall to detect and fix this.
You have just transformed your job from developer to manual spec maintainer - a clerk who has to painstakingly check everything.
Define data structures manually, ask AI to implement specific state changes. So JSON, C .h or other source files of func sigs and put prompts in there. Never tried the Agents.md monolithic definition file approach
Also I demand it stick to a limited set of processing patterns. Usually dynamic, recursive programming techniques and functions. They just make the most sense to my head and using one style I can spot check faster.
I also demand it avoid making up abstractions and stick to mathematical semantics. Unique namespaces are not relevant to software in the AI era. It's all
about using unique vectors as keys to values.
Stick to one behavior or type/object definition per file.
Only allow dependencies that are designed as libraries to begin with. There is a ton of documentation to implement a Vulkan pipeline so just do that. Don't import an entire engine like libgodot.
And for my own agent framework I added observation of my local system telemetry via common Linux files and commands. This data feeds back in to be used to generate right-sized sched_ext schedules and leverage bpf for event driven responses.
Am currently experimenting with generation of small models of my own data. A single path of images for example not the entire Pictures directory. Each small model is spun akin to a Docker container.
LLMs are monolithic (massive) zip files of the entire web. No one really asking for that. And anyone who needs it already has access to the web itself
small agents.md files are worth it, at least for holding some basic information (look at build.md to read how to build, the file structure looks like so), rather than have whatever burn double the amount of tokens searching for whatever anyways.
The funny thing is, when I got a lead position in my job, I just to do real detailed ticket descriptions, going into technical considerations and possible cross domain problems. I did it for the juniors - and to be honest - for my self, since I know if I took that ticket, from that moment to the moment I put some code down I could just forget stuff.
This was pushed back hard by management because it "took too much time to create a ticket". I fought it for some months but at the end I stopped and also really lose the ability and patience of do that. Juniors suffered, implementation took more time. Time passed.
Now, I am supposed to do the exact same thing, but even better and for yesterday.
The forcing function doesn't disappear - it shifts. When you read and critique AI-generated code carefully, you get a similar cognitive workout: Why did it structure this that way? What edge case did it miss? How does this fit the broader architecture?
The danger is treating the output as a black box. If you skip the review step and just accept whatever it produces, yes, you'll lose proficiency and accumulate debt. But if you stay engaged with the code, reading it as critically as you would a junior dev's PR, you maintain your understanding while moving faster.
The technical debt concern is valid but it's a process problem, not an inherent flaw. We solved "juniors write bad code" with code review, linting, and CI. We can solve "LLMs write inconsistent code" with the same tools - hannofcart's 10-layer static analysis stack is a good example. The LLM lies about passing checks? Pre-commit hook catches it.
Pre commit hook is definitely necessary. One thing I’ve seen a lot with Opus recently is it lying that a new linter warning or error was there before it made a change. They’ve learned from us too well!
What is the ratio of Markdown to Code with these agents? How readable is the Markdown after you've finished using it to develop your plan? How much time does it take to review code so closely vs. writing it yourself in the first place?
The first rule is an antipattern. I think describing your architecture or ANY kind of documentation for your AI is an anti-pattern and blows the context window leading to worse results, and actual more deviation.
The controlling systems are not give it more words at the start. Agentic coding needs to work in loop with dedicated context.
You need to think about how can i give as much intent as possible with as little words.
You can built a tremendous amount of custom lint rules ai never needs to read except they miss it.
Every pattern in your repo gets repeated, repo will always win over documentation and when your repo is good structured you don’t need to repeat this to AI
It’s like dev always has been, watch what has gone wrong and make sure the whole type or error can’t happen again.
it really does seem like this... also new devs are like that too: "i just copied this pattern use over here and there whats wrong?" is something i've heard over and over lol
i think languages that allow expression of "this is deprecated, use x instead" will be usefull for that too
AI gets a lot of big projects right if you give at all the tools to verify its own implementation if you can build a proper system to verify the solution it works astonishly good. Even Opus 4.6 judgement seems to be wrong most of the time on projects of my scale pre the validation layers.
In general, I prefer to do the top-level design and the big abstractions myself. I care a lot about cohesion and coupling; I like to give a lot of thought to my interfaces.
And in practice, I am happy enough that the LLM helps me to eliminate some toil, but I think you need to know when it is time to fold your cards, and leave the game. I prefer to fix small bugs in the generated code myself, than asking the agent, as it tends to go too far when fixing its own code.
The best thing about this is that AI bots will read, train on and digest the million "how to write with AI" posts that are being written right now by some of the smartest coders in the world and the next gen AI will incorporate all of this, making them ironically unnecessary.
Strange since, in practice, coding models have steadily improved without any backward movement every 3-4 months for 2 years now. It's as if there are rigorous methods of filtering and curation applied when building your training data.
>Strange since, in practice, coding models have steadily improved without any backward movement every 3-4 months for 2 years now. It's as if there are rigorous methods of filtering and curation applied when building your training data.
It's as if what I wrote implies "all other things being equal", just like any technical claim.
All other things were not equal: the architectures were tweaked, the human data set is still not exhausted, and more money and energy was thrown into their performance since it's a pre-IPO game with huge VC stakes.
We've already seen a plateau non-the-less compared to the earlier release-over-release performance improvements. Even the "without any backward movement every 3-4 months for 2 years now" is hardly arguable. Many saw a backward movement with GPT 4.1 vs 4.0, and similar issues with 4.5, for example. Even if those are isolated, they're hardly the 2 to 3.5 to 4.0 gains.
And no, there are absolutely no "rigorous methods of filtering and curation" that can separate the avalance of AI slop from useful human output - at least not without diminishing the possible training data. The problem after all is not just to tell AI from human with automated curation (that's already impossible), the problem is to have enough valuable new human output, which becomes near a losing game as all aspects of "human" domains previously useful as training input (from code to papers) are tarnished by AI output.
1. No, you dont get to fall back on the technical claim approach. Your bias in your phrasing was clear. Maybe that works for you but I won't just ignore obvious subtext and let you weasel out of this. And that's for the benefit of other readers, not you.
2. A plateau in coding performance? I don't think you even use these models for coding then if you make that claim. It is very clear models have continually improved. You can trust benchmarks to make that clear, or real world use, or better yet: both. You seem to not have the data from either.
3. No rigorous methods of filtering and curation that can separate AI slop from useful human output? Here you go:
a. Curation already works at scale.
Modern training pipelines don’t rely on “AI vs human” detection. They filter by utility signals: correctness, novelty, coherence, task success, citation integrity, and cross-source consistency. These measurable properties do correlate with downstream model performance. Models trained on smaller, higher-quality corpora consistently outperform those trained on larger, noisier ones.
b. Human-generated “valuable” data is not shrinking.
The claim assumes a fixed pool. In reality, high-value human data is expanding in areas that matter most: expert-labeled datasets, preference comparisons, multimodal demonstrations, tool-use traces, verified code with tests, and domain-expert feedback. These are explicitly created for training and are not polluted by passive AI spam.
c. Synthetic data is not a dead end—when constrained.
Empirically, filtered and goal-conditioned synthetic data (self-play, distillation, adversarial generation) improves reasoning, math, coding, and tool use. The failure mode is unfiltered synthetic recursion—not synthetic data per se. This distinction is already operationalized in production systems.
d. Training value ≠ raw text volume.
Scaling laws shifted: performance now tracks effective compute × data quality, not sheer token count. A smaller dataset with higher signal density produces better generalization than a massive, contaminated corpus. This is observed repeatedly in ablation studies.
----
Again, the above is not for you, as I believe you don't see beyond your cope (yet). It's for other readers who are intellectually curious.
>Hmm... How will it filter out those by the dumbest coders in the world?
if you know, and I know, and the guys at openai and anthropic know... not a big leap that the models will know too? many datasets are curated and labeled by humans
I mean, having a curated dataset of the works and posts of the top 200 coders in the world (at least the public ones) is not very difficult. I’m sure these articles like the one in OP will be very easy to mark as “high value training data”. I think you’re letting your bias blind you
1. Keep things small and review everything AI written, or
2. Keep things bloated and let AI do whatever it wants within the designated interface.
Initially I drew this line for API service / UI components, but it later expanded to other domains. e.g. For my hobby rust project I try to keep "trait"s to be single responsible, never overlap, easy to understand etc etc. but I never look at AI generated "impl"s as long as it passes some sensible tests and conforming the traits.
I find rust generally easier to reason about, but can't stand writing it.
The compiler works well with LLMs plenty of good tooling and LSPs.
If I'm happy with the shape of the code and I usually write the function signatures/ Module APIs. And the compiler is happy with it compiling. Usually the errors if any are logical ones I should catch in reviews.
So I focus on function, compiler focuses on correctness and LLM just does the actual writing.
Same here. I had to do a lot of being in the loop with Python, but with rust - compiler gives Claude all the information it may need and then it figures things out without me.
Writing rust scares me, but I can read it just fine. I've come up with super masochistic linting rules that claude isn't allowed to change and that has improved things quite a bit.
I wish there was a mature framework for frontend that can be configured to be as strict as rust.
Ironically, I use the time saved using agents to read technical books ferociously.
Coding agents made me really get something back from the money I pay for my O'Reilly subscription.
So, coding agents are making me a better engineer by giving me time to dive deeper into books instead of having to just read enough to do something that works under time pressure.
I found an easier way that Works For Me (TM). I describe the problem to LLM and ask it to solve it step by step, but strictly in the Ask mode, not Agent. Then I copy or even type the linws to the code. If I wouldn't write the line myself, it doesn't go in, and I iterate some more.
I do allow it to write the tests (lots of typing there), but I break them manually to see how they fail. And I do think about what the tests should cover before asking LLM to tell me (it does come up with some great ideas, but it also doesn't cover all the aspects I find important).
Great tool, but it is very easy to be led astray if you are not careful.
You still need to know the hard parts: precisely what you want to build, all domain/business knowledge questions solved, but this tool automates the rest of the coding and documentation and testing.
It's going to be a wild future for software development...
even if you check and redo after paste, you need to check for gotchas. I wish I had a nickel for every time the llm gave me a solution with a hidden limitation. assume that it violates all your unspoken assumptions, and adheres only to what you nailed down in your prompt
Same. Small units if work, iterate in it till it's right, commit it, push it, then do the next increment of work. It's how I've always worked like that, except now, I sometimes let someone else figure the exact API calls (I'm still learning react, but Claude helps get the basics in place for me). If the AI just keeps screwing up, I'll grab the wheel and do it myself. It sometimes helps me get things going, but it hasn't been a huge increase in productivity, but I'm not paying the bill so whatever.
so is the 10-20% in velocity worth the money and the process-complexity added? I'm assuming you're measuring your own velocity, not your team's, since that includes time to review and deploy etc.
Every engineering org should be pleading devs to not let AI write tests. They're awful and sometimes they literally don't even assert the code that was generated and instead assert the code in tests.
Every engineering org should be pleading devs to not let AI write code, period. They continue to routinely get stuff wrong and can't be trusted any further than you can throw them.
Ah yes, to have AI write code for you, you simply just need to, let's see ..
"Document the requirements, specifications, constraints, and architecture of your project in detail. Document your coding standards, best practices, and design patterns. Use flowcharts, UML diagrams, and other visual aids to communicate complex structures and workflows. Write pseudocode for complex algorithms and logic to guide the AI. Develop efficient debug systems for the AI to use. Build a system that collects logs from all nodes in a distributed system and provides abstracted information. Use a system that allows you to mark how thoroughly each function has been reviewed. Write property based high level specification tests yourself. Use strict linting and formatting rules to ensure code quality and consistency. Utilize path specific coding agent prompts. Provide as much high level information as practical, such as coding standards, best practices, design patterns, and specific requirements for the project. Identify and mark functions that have a high security risk, such as authentication, authorization, and data handling. Make sure that the AI is instructed to change the review state of these functions as soon as it changes a single character in the function. Developers must make sure that the status of these functions is always correct. Explore different solutions to a problem with experiments and prototypes with minimal specifications. Break down complex tasks into smaller, manageable tasks for the AI. You have to check each component or module for its adherence to the specifications and requirements."
And just like that, easy peasy, nothing to it.
As a supreme irony, the story currently on the front page directly under this one ('You are here'), makes the claim "The cost of turning written business logic into code has dropped to zero. Or, at best, near-zero." in the very first sentence.
I use it for scaffolding and often correct it for the layour I prefer. Then I use to check my code, and then scaffold in some more modules. I then connect them together.
Long as you review the code and correct it, it is no more different than using stackoverflow. A stack overflow that reads your code and helps stitch the context.
"Stack Overflow that reads your codebase" — perfect. But Stack Overflow is
stateless. Agent sessions aren't.
One session's scaffold assumes one pattern. Second session scaffold contradicts it. You reviewed both in isolation. Both looked fine. Neither knows about the other.
Reviewing AI code per-session is like proofreading individual chapters of a novel nobody's reading front to back. Each chapter is fine. The plot makes no sense.
Hi i5heu. Given that you seem to use AI tools for generating images and audio versions of your posts, I hope it is not too rude to ask: how much of the post was drafted, written or edited with AI?
The suggestions you make are all sensible but maybe a little bit generic and obvious. Asking ChatGPT to generate advice on effectively writing quality code with AI generates a lot of similar suggestions (albeit less well written).
If this was written with help of AI, I'd personally appreciate a small notice above the blog post. If not, I'd suggest to augment the post with practical examples or anecdotal experience. At the moment, the target group seems to be novice programmers rather than the typical HN reader.
i have written this text by myself except like 2 or 3 sentences which i iterated with an LLM to nail down flow and readability. I would interpret that as completely written by me.
> The suggestions you make are all sensible but maybe a little bit generic and obvious. Asking ChatGPT to generate advice on effectively writing quality code with AI generates a lot of similar suggestions (albeit less well written).
Before i wrote this text, i also asked Gemini Deep Research but for me the results where too technical and not structural or high level as i describe them here. Hence the blogpost to share what i have found works best.
> If not, I'd suggest to augment the post with practical examples or anecdotal experience. At the moment, the target group seems to be novice programmers rather than the typical HN reader.
I have pondered the idea and also wrote a few anecdotal experiences but i deleted them again because i think it is hard to nail the right balance down and it is also highly depended on the project, what renders examples a bit useless.
And i also kind of like the short and lean nature of it the last few days when i worked on the blogpost.
I might will make a few more blogposts about that, that will expand a few points.
First article about writing code with AI i can get behind 100%. Stuff i already do, stuff i've thought about doing, and at ideas i've never thought doing ("Mark code review levels" especially is a _great_ idea)
Yes this is strange. There's nothing of substance here that hasn't been repeated many times before. BUT...I did click it out of interest. So maybe it just came at an opportune lull in these types of posts and during the inflection point of 4.6 and 5.3 release. Complete guess
> Use strict linting and formatting rules to ensure code quality and consistency. This will help you and your AI to find issues early.
I've always advocated for using a linter and consistent formatting. But now I'm not so sure. What's the point? If nobody is going to bother reading the code anymore I feel like linting does not matter. I think in 10 years a software application will be very obfuscated implementation code with thousands of very solidly documented test cases and, much like compiled code, how the underlying implementation code looks or is organized won't really matter
That's the opposite. I've never read and re-read code more than i do today. The new hires generate 50 more code than they use to, and you _have_ to check it or have compounding production issues (been there, done that). And the errors can now be anywhere, when before you more or less knew what the person writing code is thinking and can understand why some errors are made. LLMs errors could hide _anywhere_, so you have to check it all.
Isn't that a losing proposition? Or do you get 50 times the value out of it too? In my experience the more verbose the code is, the less thought out it is. Lots of changes? Cool, now polish some more and come back when it's below 100 lines change, excluding tests and docs. I don't dare touch it before.
I agree, but i'm shouting at the cloud. Stuff needs to be done, it seems to work at first, so either i just abandon quality and let things rot, or i read everything and underline each time the code smell.
I too use AI, but mostly to generate scripts (the most usefull use of AI is 100-200 line scripts imho), test _cases_ (i write the test itself, the data inside is generated) and HTML/CSS/JS shenanigans (the logic i code, the presentation i'm inferior to any random guy on the internet, so i might as well use an AI). I also use it for stuff that never end in repository, for exploration/proof of concept and outside of scope tests (i like to understand how stuff work, that helps), or to summarize Powerpoint presentations so i can do actual work during 60-person "meetings" and still get the point.
That sounds like the advice of someone who doesn't actually write high-quality code. Perhaps a better title would be "how to get something better than pure slop when letting a chatbot code for you" - and then it's not bad advice I suppose. I would still avoid such code if I can help it at all.
This take is pretty uncharitable. I write high quality code, but also there's a bunch of code that could be useful, but that I don't write because it's not worth the effort. AI unlocks a lot of value in that way. And if there's one thing my 25 years as a software engineer has taught me is that while code quality and especially system architecture matter a lot, being super precious about every line of code really does not.
Don't get me wrong, I do think AI coding is pretty dangerous for those without the right expertise to harness it with the right guardrails, and I'm really worried about what it will mean for open source and SWE hiring, but I do think refusing to use AI at this point is a bit like the assembly programmer saying they'll never learn C.
Man, you are really missing out of the biggest revolution of my life.
This is the opinion of someone who has not tried to use Claude Code, in a brand new project with full permissions enabled, and with a model from the last 3 months.
People have been saying "the models from (recent timeframe) are so much better than the old ones, they solve all the problems" for years now. Since GPT-4 if not earlier. Every single time, those goalposts have shifted as soon as the next model came out. With such an abysmal track record, it's not reasonable to expect people to believe that this time the tool actually has become good and that it's not just hype.
When is the last time someone said that, motivating you to try the latest model? If it was 6 or more month ago, my reply is that the sentiment expressed was partially incorrect in the past, but it is not incorrect now. If a conspiracy theorist is always wrong about a senior citizen being killed, that does not make the senior immortal.
Where is all the amazing software and/or improvements in software quality that is supposed to be coming from this revolution?
So far the only output is the "How I use AI blogs", AI marketing blogs, more CVEs, more outages, degraded software quality, and not much of shipping anything.
Is there any examples of real products and not just anecdotes of "I'm 10x more productive!"?
I was in the same mindset until I actually took the Claude code course they offer. I was doing so much wrong.
The two main takeaways. Create a CLAUDE.md file that defines everything about the project. Have Claude feed back into the file when it makes mistakes and how to fix them.
Now it creates well structured code and production level applications. I still double check everything of course, but the level of errors is much lower.
An example application it created from a CLAUDE.md I wrote. The application reads multiple PDF's, finds the key stakeholders and related data, then generates a network graph across those files and renders it in an explorable graph in Godot.
That took 3 hours to make, test. It also supports OpenAI (lmstudio), Claude and Ollama for its LLM callouts.
What issue I can see happening is the duplication of assets in work. Instead of finding an asset someone built, people have been creating their own.
Do you find reading hard? I'm asking for examples. Why isn't anyone showing this off in blog posts. Or a youtube video or something. It's always this vague, it's faster, just trust me bro bullshit and I'm sick of it. Show me or don't reply.
So you want a video of me coding at work using AI? There are entire YouTube channels dedicated to this already. There are already copious blogs about people's AI workflows -- this very post you're commenting in is one (do you find reading hard?)
Clarify the actual thing you need to believe the technology is real or don't reply.
Its only revolutionary if you think engineers were slow before or software was not being delivered fast enough. Its revolutionary for some people sure, but everyone is in a different situation, so one man's trash can be other man's treasure. Most people are treading both paths as automation threatens their livelihood and work they loved, also still not able to understand why would people pay to companies that are actively trying to convince your employer that your job is worthless.
Even If I like this tech, I still dont want to support the companies who make it. Yet to pay a cent to these companies, still using the credits given to me by my employer.
I think there are four fundamental issues here for us...
1. There are actually less software jobs out there, with huge layoffs still going on, so software engineering as a profession doesn't seem to profit from AI.
2. The remaining engineers are expected by their employers to ship more. Even if they can manage that using AI, there will be higher pressure and higher stress on them, which makes their work less fulfilling, more prone to burnout etc.
3. Tied to the previous - this increases workism, measuring people, engineers by some output benchmark alone, treating them more like factory workers instead of expert, free-thinking individuals (often with higher education degrees). Which again degrades this profession as a whole.
3. Measuring developer productivity hasn't really been cracked before either, and still after AI, there is not a lot of real data proving that these tools actually make us more productive, whatever that may be. There is only anecdotal evidence: I did this in X time, when it would have taken me otherwise Y time - but at the same time it's well known that estimating software delivery timelines is next to impossible, meaning, the estimation of "Y" is probably flawed.
So a lot of things going on apart from "the world will surely need more software".
Yes, but who did we automate out of a job by building crappy software? Accountants are more threatened by AI than any of the software we created before, same with Lawyers, teachers. We didnt automate any physical labourers out of a job too.
It's insane! We are so far beyond gpt-3.5 and gpt-4. If you're not approaching Claude Code and other agentic coding agents with an open mind with the goal of deriving as much value from them as possible, you are missing out on super powers.
On the flip side, anyone who believes you can create quality products with these tools without actually working hard is also deluded. My productivity is insane, what I can create in a long coding session is incredible, but I am working hard the whole time, reviewing outputs, devising GOOD integration/e2e tests to actually test the system, manually testing the whole time, keeping my eyes open for stereotypically bad model behaviors like creating fallbacks, deleting code to fulfill some objective.
It's actually downright a pain in the ass and a very unpleasant experience working in this way. I remember the sheer flow state I used to get into when doing deep programming where you are so immersed in managing the states and modeling the system. The current way of programming for me doesn't seem to provide that with the models. So there are aspects of how I have programmed my whole life that I dearly miss. Hours used to fly past me without me being the wiser due to flow. Now that's no longer the case most of the times.
The article did not provide a constructive suggestion on how to write quality code, either. Nor even empirical proof in the form of quality code written by LLMs/agents via the application of those principles.
> workers who opposed the use of certain types of automated machinery due to concerns relating to worker pay and output quality... Luddites were not opposed to the use of machines per se (many were skilled operators in the textile industry); they attacked manufacturers who were trying to circumvent standard labor practices of the time.
I wonder at the end of this if it's the still worth the risk?
A lot of how I form my thoughts is driven by writing code, and seeing it on screen, running into its limitations.
Maybe it's the kind of work I'm doing, or maybe I just suck, but the code to me is a forcing mechanism into ironing out the details, and I don't get that when I'm writing a specification.
I second this. This* is the matter against which we form understanding. This here is the work at hand, our own notes, discussions we have with people, the silent walk where our brain kinda process errors and ideas .. it's always been like this since i was a kid, playing with construction toys. I never ever wanted somebody to play while I wait to evaluate if it fits my desires. Desires that often come from playing.
Outsourcing this to an LLM is similar to an airplane stall .. I just dip mentally. The stress goes away too, since I assume the LLM will get rid of the "problem" but I have no more incentives to think, create, solve anything.
Still blows my mind how different people approach some fields. I see people at work who are drooling about being able to have code made for them .. but I'm not in that group.
I'll push it back against this a little bit. I find any type of deliberative thinking to be a forcing function. I've recently been experimenting with writing very detailed specifications and prompts for an LLM to process. I find that as I go through the details, thoughts will occur to me. Things I hadn't thought about in the design will come to me. This is very much the same phenomenon when I was writing the code by hand. I don't think this is a binary either or. There are many ways to have a forcing function.
I think it's analogous to writing and refining an outline for a paper. If you keep going, you eventually end up at an outline where you can concatenate what are basically sentences together to form paragraphs. This is sort of where you are now, if you spec well you'll get decent results.
I agree, I felt this a bit. The LLM can be a modeling peer in a way. But the phase where it goes to validate / implement is also key to my brain. I need to feel the details.
My think/create/solve focus is on making my agentic coding environment produce high quality code with the least cost. Seems like a technical challenge worth playing with.
It probably helps that I have 40 years of experience with producing code the old ways, including using punch cards in middle school and learning basic on a computer with no persistent storage when I was ten.
I think I've done enough time in the trenches and deserve to play with coding agents without shame.
> I see people at work who are drooling about being able to have code made for them .. but I'm not in that group.
people seem to have a inability to predict second and third order effects
the first order effect is "I can sip a latte while the bot does my job for me"... well, great I suppose, while it lasts
but the second order effect is: unless you're in the top 10%, you will now lose your job, permanently
and the third order effect is the economy collapses as it is built on consumer spending
Alternatively, another second order effect is can't sip latte anymore because you're orchestrating 8 bots do the work and you're back to 80%-100% time saturation.
The previous second order effect is more likely. For the one orchestrating 8 bots, 7 others are not needed anymore.
Or, there is just a lot more software written as the costs drop. I think most people work with software not tailored enough for their situation..
>Or, there is just a lot more software written as the costs drop.
Yeah, no, that's wishful thinking. The companies will just opt for higher margins.
So far in my career I have always had more requests coming in than implementations going out. If I can go 3 or 10 times faster, than I will still have plenty of work. Especially for the slew of ideas that are never even considered to put towards a dev, because it's already considered to be too low value to have it even be considered to be build. Or the ideas that are so far fetched they were never considered feasible. I am not worried work will dry up.
What I believe is going to be interesting is what happens when non-engineers adopt building with agentic AI. Maybe 70 or 80% of their needs will be met without anyone else directly involved. My suspicion is that it will just create more work: making those generated apps work in a trustworthy manner, giving the agents more access to build context and make decisions, turning those one off generated apps into something maintainable, etc.
This 100%. We want MOAR!!!
Exactly this. Even if right now you, bottom level wage earning grunt, get to lighten your workload for a fleeting second, sit back and enjoy the latte it's only but a fleeting second until the capital class tighten the screws.
Most people will get laid off and made redundant and those who remain are going to have to run faster than ever to produce wealth for the capital owners.
Yea, I don’t think that will be the case. Spreadsheets simplified the work of junior finance people who did all the work by hand before. But more people work in finance now than before.
I wonder over the long term how programmers are going to maintain the proficiency to read and edit the code that the LLM produces.
There were always many mediocre engineers around, some of them even with fancy titles like "Senior," "Principal", and CTO.
We have always survived it, so probably we can also survive mediocre coders not reading the code the LLM generates for them because they are unable to see the problems that they were never able to see in their handwritten code.
Personally I planned to allocate weekly challenges to stay sharp.
Honestly it’s not that hard. I already coded less and less as part of my job as I get more senior and just didn’t have time, but I was still easy to do code reviews and fix bugs, sit down and whip out a thousand lines in a power session. Once you learn it doesn’t take much practice to maintain it. A lot of traditional coding is very inefficient. With AI it’s like we’re moving from combustion cars to EVs, the energy efficiency is night and day, for doing the same thing.
That said, the next generation may struggle, but they’ll find their way.
It’s going to be extremely difficult if PR and code reviews do not prune unnecessary functions. From what I’m experiencing now, there’s a lot of additional code that gets generated.
"That's the neat part—you don't." Eventually the workflow will be to use the LLM to interpret the LLM-generated codebase.
I don’t read or edit the code my claude code agent produces. That’s its job now. My job is to organize the process and get things done.
In this case why can’t other agents just automate your job completely ? They are capable of that. What do you bring in the process of still doing manual organization ?
I still have to tell it what to do, and often how to do it. I manage its external memory and guidelines, and review implementation plans. I’m still heavily involved in software design and test coverage.
AI is not capable yet of automating my job completely – I anticipate this will happen within two years, maybe even this year (I’m an ML researcher).
Do you mean, from your perspective, within 2 years humans won’t be able to bring anything of value to the equation in management and control ?
No, I mean that my job in its current form – as an ML researcher with a phd and 15 years of experience - will be completely automated within two years.
Is the progress of LLMs moving up abstraction layers inevitable as they gather more data from each layer? First, we fed LLMs raw text and code and now they are gathering our interactions with the LLM regarding generated code. It seems like you could then use the interactions to make a LLM that is good at prompting and fixing another LLMs generated code. Then its on to the next abstraction layer.
What you described makes sense, and it's just one of the things to try. There are lots of other research directions: online learning, more efficient learning, better loss/reward functions, better world models from training on Youtube/VR simulations/robots acting in real world, better imitation learning, curriculum learning, etc. There will undoubtedly be architectural improvements, hardware improvements, longer context windows, insights from neuroscience, etc. There is still so much to research. And there are more AI researchers now than ever. Plus current AI models already make us (AI researchers) so much more productive. But even if absolutely no further progress is made in AI research, and foundational model development stops today, there's so much improvement to be made in the tooling around the models: agentic frameworks, external memory management, better online search, better user interactions, etc. The whole LLM field is barely 5 years old.
If you want a machine (or in fact another human) to do something for you, there are two tasks you cannot delegate to them:
a) Specify what you want them to do.
b) Check if the result meets your expectations.
Does your current job include neither a nor b?
A/B happen at different abstractions levels. My abstraction level will be automated. My manager’s level will probably last another year or so.
So your assumption is that it will ultimately be the users of software themselves who will throw some every day language at an AI and it will reliably generate something that meets those users' intuitive expectations?
Yes, it will be at least as reliable as an average software engineer at an average company (probably more reliable than that), or at least as reliable as a self-driving car where a user says get me to this address, and the car does it better (statistically) than an average human driver.
I think this could work for some tasks but not for others.
We didn't invent formal languages to give commands to computers. We invented them as a tool for thinking and communicating things that are hard to express in natural language.
I doubt that we will stop thinking and I doubt that it will ever be efficient to specify tasks purely in terms of natural language.
One of my first jobs as a software engineer was for a bank (~30 years ago). This bank manager wasn't a man of many words. He just handed us an Excel sheet as a specification for what he wanted us to implement.
My job right now is to translate natural English statements from my bosses/colleagues into natural English instructions for Claude. Yes, it takes skill and experience to do this effectively. But I don't see any reasons Gemini 4, Opus 5 or GPT-6 won't be able to do this just as well as I do.
What are you going to do for work in 2 years?
I have enough savings for a few years, so I might just move to a lower COL area, and wait it out. Hopefully after the initial chaos period things will improve.
For someone at your position with your experience it’s quite depressing that your job is going to be automated. I feel quite anxious when I see young generations in my country that say themselves they are lazy about learning new things. The next generation will be useless to capitalist societies, in a sense that they won’t be able to bring value through administrative or white collar work. I hope some areas of the industry will move slowly toward AI
simonw alert!!!
Actually for me it was the opposite: before I wasn't able to play around and experiment in my free time that much, because I didn't have enough energy left to actualize the thoughts and ideas I have since I have a day job.
Now, since the bottleneck of moving the fingers to write code has gone down, I actually started to enjoy doing side projects. The mental stress from writing code has gone down drastically with Claude Code, and I feel the urge to create more nowadays!
you have a point.. i'm still confused about how this will affect jobs, markets
in a way a personal project is different from a job duty, here you're exploring, less if no deadline.. at work if I feel the llm is doing everything and I don't really master, i risk my job and my skills rot.
>I see people at work who are drooling about being able to have code made for them .. but I'm not in that group.
+100 for this.
> I see people at work who are drooling about being able to have code made for them
These people just drool at being able to have work done for them to begin with. Are you sure it is just "code"?
> I see people at work who are drooling about being able to have code made for them .. but I'm not in that group.
In my circles see some overlap with the people who are like: "Done! Let's move on" and don't worry about production bugs, etc. "We'll fix it later".
I've always stressed out about introducing bugs and want to avoid firefighting (even in orgs where that's the way to get noticed).
Too much leaning on coding tools and agents feels to sketchy to someone like me right now (maybe always tbh)
Everything you have said here is completely true, except for "not in that group": the cost-benefit analysis clearly favors letting these tools rip, even despite the drawbacks.
Maybe.
But it's also likely that these tools will produce mountains of unmaintainable code and people will get buried by the technical debt. It kind of strikes me as similar to the hubris of calling the Titanic "unsinkable." It's an untested claim with potentially disastrous consequences.
> But it's also likely that these tools will produce mountains of unmaintainable code and people will get buried by the technical debt.
It's not just likely, but it's guaranteed to happen if you're not keeping an eye on it. So much so, that it's really reinforced my existing prejudice towards typed and compiled languages to reduce some of the checking you need to do.
Using an agent with a dynamic language feels very YOLO to me. I guess you can somewhat compensate with reams of tests though. (which begs the question, is the dynamic language still saving you time?)
Companies aren't evaluating on "keeping an eye on technical debt", but then ARE directly evaluating on whether you use AI tools.
Meanwhile they are hollowing out work forces based on those metrics.
If we make doing the right thing career limiting this all gets rather messy rather quickly.
> If we make doing the right thing career limiting this all gets rather messy rather quickly.
This has already happened. The gold rush brogrammers have taken over.
Careers are over. Company loyalty is a relic. Now it's a matter of adapting quickly to earn enough to survive.
Tests make me faster. Dynamic or not feels irrelevant when I consider how much slower I’d be without the fast feedback loop of tests.
You can (and probably should) still do tests, but there's an entire class of errors you know can't happen, so you need far less tests, focusing only on business logic for the most part.
Static type checking is even faster than running the code. It doesn't catch everything, but if finding a type error in a fast test is good, then finding it before running any tests seems like it would be even better.
I can provide evidence for your claim. The technical debt can easily snowball if the review process is not stringent enough to keep out unnecessary functions.
> the cost-benefit analysis clearly favors letting these tools rip
Does it? I have yet to see any evidence that they are a net win in terms of productivity. It seems to just be a feeling that it's more efficient.
Oh I'm well aware of this. I admitted defeat in a way.. I can't compete. I'm just at loss, and unless LLM stall and break for some reason (ai bubble, enshittification..) I don't see a future for me in "software" in a few years.
Somehow I appreciate this type of attitude more than the one which reflects total denial of the current trajectory. Fervent denial and AI trash-talking being maybe the single most dominant sentiment on HN over the last year, by all means interspersed with a fair amount of amazement at our new toys.
But it is sad if good programmers should loose sight of the opportunities the future will bring (future as in the next few decades). If anything, software expertise is likely to be one of the most sought-after skills - only a slightly different kind of skill than churning out LOCs on a keyboard faster than the next person: People who can harness the LLMs, design prompts at the right abstraction level, verify the code produced, understand when someone has injected malware, etc. These skills will be extremely valuable in the short to medium term AFAICS.
But ultimately we will obviously become obsolete if nothing (really) catastrophic happens, but when that happens then likely all human labor will be obsolete too, and society will need to be organized differently than exchanging labor for money for means of sustenance.
I get crazy over the 'engineer are not paid to write loc', nobody is sad because they don't have to type anymore. My two issues are it levels the delivery game, for the average web app, anybody can now output something acceptable, and then it doesn't help me conceptualize solution better, so I revert to letting it produce stuff that is not maleable enough.
I wonder about who "anybody can now output something acceptable" will hit most - engineers or software entrepreneurs.
Any implementation moat around rapid prototyping, and any fundraising moat around hiring a team of 10 to knock out your first few versions, seems gone now. Trying to sell MVP-tier software is real hard when a bunch of your potential customers will just think "thanks for the idea, I'll just make my own."
The crunch for engineers, on the other hand, seems like that even if engineers are needed to "orchestrate the agents" and manage everything, there could be a feature-velocity barrier for the software that you can still sell (either internally or externally). Changing stuff more rapidly can quickly hit a point of limited ROI if users can't adjust, or are slowed by constant tooling/workflow churn. So at some point (for the first time in many engineers' career, probably) you'll probably see product say "ok even though we built everything we want to test, we can't roll it all out at once!". But maybe what is learned from starting to roll those things out will necessitate more changes continually that will need some level of staffing still. Or maybe cheaper code just means ever-more-specialized workflows instead of pushing users to one-size-fits-all tooling.
In both of those cases the biggest challenge seems to be "how do you keep it from toppling down over time" which has been the biggest unsolved problem in consumer software development for decades. There's a prominent crowd right now saying "the agents will just manage it by continuing to hack on everything new until all the old stuff is stable too" but I'm not sure that's entirely realistic. Maybe the valuable engineering skills will be putting in the right guardrails to make sure that behavioral verification of the code is a tractable problem. Or maybe the agents will do that too. But right now, like you say, I haven't found particularly good results in conceptualizing better solutions from the current tools.
> your potential customers will just think "thanks for the idea, I'll just make my own."
yeah, and i'm surprised nobody talks about this much. prompting is not that hard, and some non software people are smart enough to absorb the necessary details (especially since the llm can tutor them on the way) and then let the loop produce the MVP.
> Or maybe cheaper code just means ever-more-specialized workflows instead of pushing users to one-size-fits-all tooling.
Interesting thought
If the world comes to that it will be absolutely catastrophic, and it’s a failure of grappling with the implications that many of the executives of AI companies think you can paper over the social upheaval with some UBI. There will be no controlling what happens, and you don’t even need to believe in some malicious autonomous AI to see that.
The future is either a language model trained on AI code bloats and the ways to optimize the bloat away
OR,
something like Mercor, currently getting paid really well by Meta, OpenAI, Anthropic and Gemini to pay very smart humans really well to proof language model outputs.
Yep, its a rather depressing realization isnt it. Oh well, life moves on i suppose.
I think we realistically have a few years of runway left though. Adoption is always slow outside of the far right of the bell curve.
i'm sorry if I pulled everybody down .. but it's been many months since gemini and claude became solid tools, and regularly i have this strong gut feeling. i tried reevaluating my perception of my work, goals, value .. but i keep going back to nope.
After a multi-decade career that spanned what is rapidly seeming like the golden age of software development, I have two emotions: first gratefulness; second a mixture of resignation, maudlin reflection, and bitterness that I am fighting hard to resist.
As someone who’s always wanted to “get home and code something on my own”, I do have a glimmer of hope that I wonder if others share. I’ve worked extensively with Claude and there’s no question I am now a high velocity “builder” and my broad experience has some value here. I am sad that I won’t be able to deeply look at all the code I am producing, but I am making sure the LLM and I structure things so that I could eventually dig in to modules if needed (unlikely to happen I suppose).
Anyway, my hope/question: if I embrace my new role as fast system builder and I am creative in producing systems that solve real problems “first”, is there a path to making that a career (I.e. 4 friends and I cranking out real production software that’s filling a real niche)? There must be some way for this to succeed —- I am not yet buying the “everything will be instantly copyable and so any solution is instantly commodity” argument. If that’s true, then there is no hope. I am still in shape, though, so going pro in pickleball is always an option, ha ha.
Unfortunately you aren't a high velocity builder. The velocity curve has now shifted and everyone having Claude blast out loc after loc is now a high velocity builder. And when everyone is a high velocity builder...nobody is.
“And when everyone’s super, no one will be”.
Fair point, but my hope is that the creativity involved in deciding what to build, with the choice informed by engineering experience (the project/value will not be obvious to everyone) will allow differentiation.
"creativity involved in deciding what to build, with the choice informed by engineering experience (the project/value will not be obvious to everyone) will allow differentiation."
How? Anyone upon seeing your digital product can just prompt the same thing in no time. If you can prompt it, I can prompt it and so can a million other people.
Nobody whether an individual or business holds any uniqueness or advantage to themselves. All careers and skill sets are leveled and worthless. Implementation skills are worthless. Creativity is worthless.
The only valuable thing is data.
Agree on data value, but as mentioned above I am not yet buying the “everything will be instantly copyable and so any solution is instantly commodity” argument … crud web-app sure, something with significant back-end complexity or a multi-service systems level solution, not so much. Perhaps optimistic, admittedly. Cheers.
I hear you. And maybe you're right. Maybe I'm deluding myself, but: when I look at my skilled colleagues who vibecode, I can't understand how this is sustainable. They're smart people, but they've clearly turned off. They can't answer non-trivial questions about the details of the stuff they (vibe-)delivered without asking the LLM that wrote it. Whoever uses the code downstream aren't gonna stand (or pay!) for this long-term! And the skills of the (vibe-)authors will rapidly disappear.
Maybe I'm just as naive as those who said that photographs lack the soul of paintings. But I'm not 100% convinced we're done for yet, if what you're actually selling is thinking, reasoning and understanding.
The difference with a purely still photograph is that code is a functional encoding of an intention. Code of an LLM could be perfect and still not encode the perfect intention of the product. I’ve seen that in many occasions. Many people don’t understand what code really is about and think they have a printer toy now and we don’t have to use pencils. That’s not at all the same thing. Code is intention, logic, specific use case all at once. With a non deterministic system and vague prompting there will be misinterpreted intentions from LLM because the model makes decisions to move forward. The problem is the scale of it, we’re not talking about 1000 loc. In a month you can generate millions of loc, in a year hundreds of millions of loc.
Some will have to crash and burn their company before they realize that no human at all in the loop is a non sense. Let them touch fire and make up their mind I guess.
> Code is intention, logic, specific use case all at once. With a non deterministic system and vague prompting there will be misinterpreted intentions from LLM because the model makes decisions to move forward. The problem is the scale of it, we’re not talking about 1000 loc. In a month you can generate millions of loc, in a year hundreds of millions of loc.
People are also non deterministic. When I delegate work to team of five or six mid level developers or God forbid outsourced developers, I’m going to have to check and review their work too.
It’s been over a decade that my vision/responsibility could be carried out by just my own two hands and be done on time within 40 hours a week - until LLMs
Ofc people are non deterministic. But usually we expect machines to be. That’s why we trust them blindly and don’t check the calculations. We review people’s work all the time though. Here people will stop review machine LLM code as it’s kind of a source of truth like in other areas. That’s my point, reviewing code takes time and even more time when no human wrote it. It’s a dangerous path to stop reviews because of trust in the machine now that the machine is just kind of like humans, non deterministic.
No one who has any knowledge or who has ever used an LLM expects determinism.
And there are no computer professionals who haven’t heard about hallucinations.
Reviewing whether the code meets requirements through manual and automated tests - and that’s all I cared about when I had a team of 8 under me - is the same regardless. I wasn’t checking whether John used a for loop or while loop in between my customer meetings and meetings with the CTO. I definitely wasn’t checking the SOQL (not a typo) of the Salesforce consultants we hired. I was testing inputs and outputs and UX.
Having a team of 8 people producing code is manageable. Having an AI with 8 agents that write code all day long is not the same volume it can generate more code in a day that one person can review in a week. What you say is that, product teams will prompt what they want to a framework, the framework will take care of spec analysis, development, reviews, compliance with spec. Product teams with QA will make sure the delivery is functionally correct. No humans need to make sure of anything code related. What we don’t know yet is, does AI will still produce solid code trough the years because it’s all statistical analysis and with the volume of millions of loc, refactoring needed, data migrations etc what will happen ?
For context, I just started using coding agents - codex CLI and Claude code in October. Once I saw that you had to be billed by use, I’m not using my own money for it when it’s for a company.
Two things changed - Codex CLI now lets you use it with your $20 a month subscription and I have never run into quota issues with it and my employer signed up for the enterprise vs of Claude and we each have an $800 a month allowances
My argument though is “why should I care about the code?” for the most part. If I were outsourcing a project or delegating it to a team lead, I would be asking high level architectural, security and scalability questions.
AI generated the code, AI maintains the code. I am concerned about abstractions and architecture.
You shouldn’t have to maintain or refactor “millions of lines of code”, if your code is well modularized with clean interfaces, making a change for $x7 may mean making a change for $x1…$x6. But you still should be working locally in one module at the time. You should do the same for the benefit of coders. Heck my little 5 week project has three independently deployable repos in a root folder. My root Agents file just has a summary of how all three relate via a clean interface.
In the project I am working on now, besides “does it meet the requirements”, I care about security, scalability, concurrency, user experience for the end user, user experience for the operations folks when they need to make config changes, and user experience for any developers who have to make changes long after I’m off this project. I haven’t looked at a single line of code - besides the CloudFormation templates. But I can answer any architectural question about any of it. The architecture and abstractions were designed by me and dictated to the agents
On this particular project, on the coding level, there is absolutely nothing that application code like this can do that could be insecure except hypothetically embed AWS credentials into the code. But it can’t do that either since it doesn’t have access to it [1].
In this case security posture comes from the architecture - S3 block public access, well scoped IAM roles, not running “in a VPC”. Things I am checking in the infrastructure as code and I was very specific about.
The user experience has to come from design and checking manually.
I mentioned earlier that my first stab it scaled poorly. This was caused by my design and I suspected it would beforehand. But building the first version was so fast because of AI tools, I felt no pain in going with my more architecturally complicated plan B and throwing the first version away. I wouldn’t have known that by looking at the code. The code was fine it was the underlying AWS service. I could only know that by throwing 100K documents at it instead of 1000.
I designed a concurrent locking mechanism that had a subtle flaw. Throwing the code into ChatGPT into thinking mode, it immediately found it. I might have been better off just to tell the coding agents “design a locking mechanism for $x” instead of detailing it.
Even maintainability was helped because I knew I or anyone else who touched it was probably going to be using an LLM. From the get go I threw the initial contract, the discovery sessions transcripts, the design diagrams, the review of the design diagrams, my project plan and breakdown into ChatGPT and told it to render a detailed markdown file of everything - that was the beginning of my AGENTS.md file.
I asked both Codex and Claude to log everything I was doing and my decisions into separate markdown files.
Any new developer could come into my repo, fire up Claude and it wouldn’t just know what was coded, it would have full context of the project from the initial contract through to the delivery
[1] code running on AWS never explicitly has to worry about AWS credentials , the SDKs can find the information by themselves by using the credentials of the IAM role attached to the EC2 instance, Lambda, Docker container, etc.
Even locally you should be getting temporary credentials that are assigned to environment variables that the SDK retrieved automatically.
There are so many types of requirements though. Security is one, performance is another. No one has cared about while/for for a long time.
Okay - and the person ultimately leading the team is still responsibility for it whether you are delegating to more junior developers or AI. You’re still reviewing someone else’s code based on your specs
People are indeed not deterministic. But they are accountable. In the legal sense, of course, but more importantly, in an interpersonal sense.
Perhaps outsourcing is a good analogy. But in that case I'd call it outsourcing without accountability. LLMs feel more like an infinite chain of outsourcing.
As a former tech lead and now staff consultant who leads cloud implementations + app dev, I am ultimately responsible for making sure that projects are done on time, on budget and meets requirements. My manager nor the customer would allow me to say it’s one of my team members fault that something wasn’t done correctly any more than I could say don’t blame me blame Codex.
I’ve said repeatedly over the past couple of days that if a web component was done by someone else, it might as well have been created by Claude, I haven’t done web development in a decade. If something isn’t right or I need modifications I’m going to either have to Slack the web developer or type a message to Claude.
I have this nagging feeling I’m more and more skimming text, not just what the LLMs output, but all type of texts. I’m afraid people will get too lazy to read, when the LLM is almost always right. Maybe it’s a silly thought. I hope!
This is my fear too.
People will say "oh, it's the same as when the printing press came, people were afraid we'd get lazy from not copying text by hand", or any of a myriad of other innovations that made our lives easier. I think this time it's different though, because we're talking about offloading the very essence of humanity – thinking. Sure, getting too lazy to walk after cars became widespread was detrimental to our health, but if we get too lazy to think, what are we?
there are some youtube videos about the topic, be it pupil in high school addicted to llms, or adults losing skills, and not dev only, society is starting to see strange effects
Can you provide links to these videos?
This one is in french (hope you don't mind), https://youtu.be/4xq6bVbS-Pw?t=534 mentions the issues for students and other cognitive issues.
I feel the same. And I expect even a lot of the early adopters and AI enthusiasts are going to find themselves as the short end of the stick sooner than later.
"Oops I automated myself out a job".
I've already seen this play out. The lazies in our floor were all crazy about AI because they could finally work few and finish their tasks. Until they realized that they were visibly replaceable now. The motto in team chats is "we'll lie about the productivity gains to management, just say 10% but with lots of caretaking" now
Imagine everyone who is in less technical or skilled domains.
I can't help but resist this line of thinking as a result. If the end is nigh for us, it's nigh for everyone else too. Imagine the droves of less technical workers in the workforce who will be unseated before software engineers. I don't think it is tenable for every worker in the first world to become replaced by a computer. If an attempt at this were to occur, those smart unemployed people would be a real pain in the ass for the oligarchs.
I feel the same.
Frankly, I am not sure there is a place in the world at all for me in ten years.
I think the future might just be a big enough garden to keep me fed while I wait for lack of healthcare access to put me out of my misery.
I am glad I am not younger.
Yup. The majority of this website is going to find out they were grossly overpaid for a long time.
So why havent you been fired already?
.......
gemini has only been deployed in the corp this year, but the expectations are now higher (doubled). i'll report by the end of the year..
That's also how I feel.
I think you have every right to doubt those telling us that they run 5 agents to generate a new SAAS-product while they are sipping latté in a bar. To work like that I believe you'll have to let go of really digging into the code, which in my experience is needed if want good quality.
Yet I think coding agents can be quite a useful help for some of the trivial, but time consuming chores.
For instance I find them quite good at writing tests. I still have to tweak the tests and make sure that they do as they say, but overall the process is faster IMO.
They are also quite good at brute-forcing some issue with a certain configuration in a dark corner of your android manifest. Just know that they WILL find a solution even if there is none, so keep them on a leash!
Today I used Claude for bringing a project I abandoned 5 years ago up to speed. It's still at work in progress, but the task seemed insurmountable (in my limited spare time) without AI, now it feels like I'm half-way there in 2-3 hours.
I think we really need to have a serious think of what is "good quality" in the age of coding agents. A lot of the effort we put into maintaining quality has to do with maintainability, readability etc. But is it relevant if the code isn't for humans? What is good for a human is not what is good for an AI necessarily (not to say there is no overlap). I think there are clearly measurable things we can agree still apply around bugs, security etc, but I think there are also going to be some things we need to just let go of.
>But is it relevant if the code isn't for humans?
The implications to your statement seems to me that is: "you'll never have to directly care about it yourself, so why do you care about it?". Unless you were talking about the codebase in a user-application relationship which in this case feel free to ignore the rest of my post.
I don't believe that the code will become an implementation detail, ever. When all you do is ship an MVP to demonstrate what you're building then no one cares, before or after LLM assistance. But any codebase that lives more than a year and serves real users while generating revenue deserves to have engineers who knows what's happening beyond authoring markdown instructions to multiple agents.
Your claim seems to push us towards a territory where externalizing out thought processes to a third party is the best possible outcome for all parties, because the models will only get better and stay just as affordable.
I will respond to that by pointing out that, models that will ultimately be flawless in code generation will worth a fortune in terms of adding value, and any corporation that will win the arms race will be actually killing themselves by not raising the cost of access to their services by a metric ton. This is because there will be few LLM providers that actually worth it by then, and because oligopoly is a thing.
So no. I don't expect that we'll ever reach a point where the average person will be "speaking forth" software the same way they post on Reddit, without paying cancer treatment levels of money.
But even if it's actually affordable... Why would I ever want to use your app instead of just asking an LLM to make me one from scratch? No one seems to think about that.
i've been building agent tooling for a while and this is the question i keep coming back to. the actual failure mode isn't messy code, agents produce reasonably clean, well-typed output these days. it's that the code confidently solves a different problem than what you intended. i've had an agent refactor an auth flow that passed every test but silently dropped a token refresh check because it "simplified" the logic. clean code, good types, tests green, security hole. so for me "quality" has shifted from cyclomatic complexity and readability scores to "does the output behaviour match the specification across edge cases, including the ones i didn't enumerate." that's fundamentally an evaluation problem, not a linting problem.
This is where I think its going, it feels that in the end we will end up with an "llm" language, one that is more suited to how an llm works and less human.
You can’t drop anything as long as a programmer is expected to edit the source code directly. Good luck investigating a bug when the code is unclear semantically, or updating a piece correctly when you’re not really sure it’s the only instance.
I think that's the question. Is a programmer expected to ever touch the source code? Or will AI -- and AI alone -- update the code that it generated?
Not entirely unlike other code generation mechanisms, such as tools for generating HTML based on a graphical design. A human could edit that, but it may not have been the intent. The intent was that, if you want a change, go back to the GUI editor and regenerate the HTML.
> Not entirely unlike other code generation mechanisms, such as tools for generating HTML based on a graphical design. A human could edit that, but it may not have been the intent. The intent was that, if you want a change, go back to the GUI editor and regenerate the HTML.
We largely moved back away from "work in a graphic tool then spit out HTML from it" because it wasn't robust for the level of change/iteration pace, this wasn't exactly my domain but IIRC there were especially a lot of problems around "small-looking changes are now surprisingly big changes in the generated output that have a large blast radius in terms of the other things (like interactivity) we've added in."
Any time you do a refactor that changes contract boundaries between functions/objects/models/whatever, and you have to update the tests to reflect this, you have a big risk of your new tests not covering exactly the same set of component interactions that your old tests did. LLM's don't change this. They can iterate until the tests are green, but certain changes will require changing the tests, and now "iterating until the tests are green" could be resolved by changing the tests in a way that subtly breaks surprising user-facing things.
The value of good design in software is having boundaries aligned with future desires (obviously this is never perfect foresight) to minimize that risk. And that's the scary thing to myself about not even reading the code.
So like we went from assembler to higher level programming languages, we will now move to specifications for LLMs? Interesting thought... Maybe, once the "compilers" get good enough, but for mission critical systems they are not nearly good enough yet.
Right. I work in aerospace software, and I do not know if this option would ever be on the table. It certainly isn't now.
So I think this question needs to be asked in the context of particular projects, not as an industry-wide yes or no answer. Does your particular project still need humans involved at the code level? Even just for review? If so, then you probably ought to retain human-oriented software design and coding techniques. If not, then, whatever. Doesn't matter. Aim for whatever efficiency metric you like.
Not everyone works in aerospace engineering, though.
I would guess that >90% of all web crud can already be done better by an LLM managed by a decent developer, than purely by the developer himself.
Then again, would anyone have guessed we’d even be seriously discussing this topic 10, 20, 40 years ago?
Maybe. This book from 1990
https://mitpress.mit.edu/9780262526401/artificial-intelligen...
envisions a future of AI assistance that looks not too far off from today.
It’s also pretty close to Steve Jobs initial vision of computing in the future (https://stevejobsarchive.com/stories/objects-of-our-life, 1983) but my point is that whatever it is we call AI now became reality so much faster than anyone really saw coming. Even if the pace slows down, and it didn’t yet, things are improving so massively all the time that the world can’t keep up changing to accommodate.
This is exactly what is happening from a levels of abstraction standpoint.
The difference being that compilers and related tools are deterministic, and we can manage the outputs using mathematical proof of correctness.
The LLM's driving this new abstraction layer are another beast entirely.
> I think you have every right to doubt those telling us that they run 5 agents to generate a new SAAS-product while they are sipping latté in a bar. To work like that I believe you'll have to let go of really digging into the code, which in my experience is needed if want good quality.
Also we live in a capitalist society. The boss will soon ask: "Why the fuck am I paying you to sip a latte in a bar? While am machine does your work? Use all your time to make money for me, or you're fired."
AI just means more output will be expected of you, and they'll keep pushing you to work as hard as you can.
> AI just means more output will be expected of you, and they'll keep pushing you to work as hard as you can.
That’s a bit too cynical for me. After all, yes, your boss is not paying you for sipping lattes, but for producing value for the company. If there is a tool that maximises your output, why wouldn’t he want you to use that to great efficiency?
Put differently, would a carpenter shop accept employees rejecting the power saw in favour of a hand saw to retain their artisanal capability?
> why wouldn’t he want you to use that to great efficiency
Because I deny that? It's not fun for me.
> would a carpenter shop accept employees rejecting the power saw in favour of a hand saw to retain their artisanal capability?
Why not? If that makes enough money to keep going.
You might argue that in theoretical ideal market companies who're not utilizing every possible trick to improve productivity (including AI) will lose competition, but let's be real, a lot of companies are horribly inefficient and that does not make them bankrupt. The world of producing software is complicated.
I know that I deliver. When I'm asked to write a code, I deliver it and I responsible for it. I enjoy the process and I can support this code. I can't deliver with AI. I don't know what it'll generate. I don't know how much time would it take to iterate to the result that I precisely want. So I can't longer be responsible for my own output. Or I'd spend more time baby-sitting AI than it would take me to write the code. That's my position. Maybe I'm wrong, they'll fire me and I'll retire, who knows. AI hype is real and my boss often copy&pasting ChatGPT asking me to argue with it. That's super stupid and irritating.
I can totally relate to your experience.
I started this career because I liked writing code. I no longer write a lot of code as a lead, but I use writing code to learn, to gain a deeper understanding of the problem domain etc. I'm not the type who wants to write specs for every method and service but rather explore and discover and draft and refactor by... well, coding. I'm amazed at creating and reading beautiful, stylish, working code that tells a story.
If that's taken away, I'm not sure how I could retain my interest in this profession. Maybe I'll need to find something else, but after almost a decade this will be a hard shift.
> Because I deny that? It's not fun for me.
I totally emphasise as a fellow developer, but I doubt you realise what an incredibly privileged position it is to just refuse working if you don't have fun doing it. And it doesn't really make for a convincing argument to keep you employed either.
> Why not? If that makes enough money to keep going.
If all other competing carpenters use power tools, you're going to loose contracts. We've had a few incredibly easy decades as software developers where market pressure wasn't really a thing, but that is about to change when the cost of producing code drops considerably.
> You might argue that in theoretical ideal market companies who're not utilizing every possible trick to improve productivity (including AI) will lose competition […]
You're moving the goalposts here. We're not talking about wringing every last drop of efficiency out of employees. We're talking about businesses not tolerating paying for licenses for AI agents to enable developers to sip Lattés while their computer does their job. That's a fundamentally different proposition.
> That’s a bit too cynical for me. After all, yes, your boss is not paying you for sipping lattes, but for producing value for the company. If there is a tool that maximises your output, why wouldn’t he want you to use that to great efficiency?
Sitting in a cafe enjoying a latte is not "producing value for the company." If having "5 agents to generate a new SAAS-product" matches your non-AI capacity and gives you enough free time to relax in a cafe, he's going to want to you run 50 agents generating 5 new SAAS products, until you hit your capacity.
If he doesn't need 5 new SAAS products, just one, then he's going to fire you or other members of your team.
Think of it this way: you're a piece of equipment to your boss, and every moment he lets you sit idle (on the clock) is money lost. He wants to run that piece of equipment as hard as he can, to maximize his profit.
That's labor under capitalism.
If the power saw ran itself without any oversight, the carpenter shop wouldn't accept any type of employees.
But that’s the exact opposite of what the GP was arguing; you will be expected to stick with the agent more, not less.
You or someone else might be expected. The rest will just be expected to be fired.
I still do this, but when I'm reviewing what's been written and / or testing what's been built.
How I see it is we've reverted back to a heavier spec type approach, however the turn around time is so fast with agents that it still can feel very iterative simply because the cost of bailing on an approach is so minimal. I treat the spec (and tests when applicable) as the real work now. I front load as much as I can into the spec, but I also iterate constantly. I often completely bail on a feature or the overall approach to a feature as I discover (with the agent) that I'm just not happy with the gotchas that come to light.
AI agents to me are a tool. An accelerator. I think there are people who've figured out a more vibey approach that works for them, but for now at least, my approach is to review and think about everything we're producing, which forms my thoughts as we go.
>but the code to me is a forcing mechanism into ironing out the details, and I don't get that when I'm writing a specification.
This is so on point. The spec as code people try again and again. But reality always punches holes in their spec.
A spec that wasn't exercised in code, is like a drawing of a car, no matter how detailed that drawing is, you can't drive it, and it hides 90% of the complexity.
To me the value of LLMs is not so much in the code they write. They're usually to verbose, start building weird things when you don't constantly micromanage them.
But you can ask very broad questions, iteratively refine the answer, critique what you don't like. They're good as a sounding board.
I love using LLMs as well as rubber ducks - what does this piece of code do? How would you do X with Y? etc.
The problem is that this spec-driven philosophy (or hype, or mirage...) would lead to code being entirely deprecated, at least according to its proponents. They say that using LLMs as advisors is already outdated, we should be doing fully agentic coding and just nudge the LLM etc. since we're losing out on 'productivity'.
>They say that using LLMs as advisors is already outdated, we should be doing fully agentic coding and just nudge the LLM etc. since we're losing out on 'productivity'.
As long as "they" are people that either profit from FOMO or bad developers that still don't produce better software than before, I'm ok ignoring the noise.
Historically software engineering has been seen as "assembly line" work by a lot of people (see all the efforts to outsource it through spec handoffs and waterfall through the years) but been implemented in practice as design-as-you-build (nobody anticipates all the questions or edge cases in advance, software specs are often an order of magnitude simpler than the actual number of branches in the code).
For mission-critical applications I wonder if making "writing the actual code" so much cheaper means that it would make more sense to do more formal design up front instead, when you no longer have a human directly in the loop during the writing of the code to think about those nasty pops-up-on-the-fly decisions.
> software specs are often an order of magnitude simpler than the actual number of branches in the code
Love this! Be it design specs or a mock from the designer. So many unaccounted for decisions. Good devs will solve many on their own, uplevel when needed, and provide options.
And absolutely it means more design up front. And without human in the direct loop, maybe people won’t skimp on this!
I also second this. I find that I write better by hand, although I work on niche applications it’s not really standard crud or react apps. I use LLMs in the same way i used to used stack overflow, if I go much farther to automate my work than that I spend more time on cleanup compared to if I just write code myself.
Sometimes the AI does weird stuff too. I wrote a texture projection for a nonstandard geometric primitive, the projection used some math that was valid only for local regions… long story. Claude kept on wanting to rewrite the function to what it thought was correct (it was not) even when I directed to non related tasks. Super annoying. I ended up wrapping the function in comments telling it to f#=% off before it would leave it alone.
i've asked an ai to plan and setup some larger non straight forwards changes/features/refactorings but it usually devolves into burning tokens and me clicking the 'allow' button and re-clarifying over and over when it keeps trying to confirm the build works etc...
when i'm stuck though, or when im curious of some solution it usually opens the way to finish the work similar to stack overflow
Exactly. 30 years ago a mathematician I knew said to me: "The one thing that you can say for programming is that it forces you to be precise."
We vibe around a lot in our heads and that's great. But it's really refreshing, every so often, to be where the rubber meets the road.
Using AI or writing your own code isn't an xor thing. You can still write the code but have a coding assistant or something an alt/cmd-tab away. I enjoy writing code, it relaxes me so that's what I do but when I need to look something up or i'm not clear on the syntax for some particular operation instead of tabbing to a browser and google.com I tab to the agent and ask it to take a look. For me, this is especially helpful for CSS and UI because I really suck at and dislike that part of development.
I also use these things to just plan out an approach. You can use plan mode for yourself to get an idea of the steps required and then ask the agent to write it to a file. Pull up the file and then go do it yourself.
In 1987 when I first started coding, I would either write my first attempt in BASIC and see it was too slow and rewrite parts in assembly or I would know that I had to write what I wanted from the get go in assembly because the functionality wasn’t exposed at all in BASIC (using the second 64K of memory or using double hires graphics).
This past week, I spent a couple of days modifying a web solution written by someone else + converting it from a Terraform based deployment to CloudFormation using Codex - without looking at the code as someone who hasn’t done front in development in a decade - I verified the functionality.
More relevantly but related, I spent a couple of hours thinking through an architecture - cloud + an Amazon managed service + infrastructure as code + actual coding, diagramming it, labeling it , and thinking about the breakdown and phases to get it done. I put all of the requirements - that I would have done anyway - into a markdown file and told Claude and Codex to mark off items as I tested each item and summarize what it did.
Looking at the amount of work, between modifying the web front end and the new work, it would have taken two weeks with another developer helping me before AI based coding. It took me three or four days by myself.
The real kicker though is while it worked as expected for a couple of hundred documents, it fell completely to its knees when I threw 20x documents into the system. Before LLMs, this would have made me look completely incompetent telling the customer I now wasted two weeks worth of time and 2 other resources.
Now, I just went back to the literal drawing board, rearchitected it, did all of the things with code that the managed services abstracted away with a few tweaks, created a new mark down file and was done in a day. That rework would have taken me a week by itself. I knew the theory behind what the managed service was doing. But in practice I had never done it.
It’s been over a decade where I was responsable for a delivery that I could do by myself without delegating to other people or that was simple enough that I wouldn’t start with a design document for my own benefit. Now within the past year, I can take on larger projects by myself without the coordination/“mythical man Month” overhead.
I can also in a moment of exasperation say to Codex “what you did was an over complicated stupid mess, rethink your implementation from first principles” without getting reported to HR.
There is also a lot of nice to have gold plating that I will do now knowing that it will be a lot faster
That's because many developers are used to working like this.
With AI, the correct approach is to think more like a software architect.
Learning to plan things out in your head upfront without to figure things out while coding requires a mindset shift, but is important to work effectively with the new tools.
To some this comes naturally, for others it is very hard.
> Learning to plan things out in your head
I dont think any complex plan should be planned in your head. But drawing diagrams, sketching components, listing pros and cons, 100%. Not jumping directly into coding might look more like jumping into spec writing a poc
Maintaining a 'mental RAM Cache' is a powerful tool to understanding the system as a whole on a deep and intuitive level, even if you can only 'render' sections at a time. The bigger it is the more you can keep track of to be able to foresee interactions between distant pieces.
It shouldn't be your only source of a plan as you'd likely wind up dropping something, but figuring out how to jiggle things around before getting it 'on paper' is something I've found helpful.
Following the RAM analogy, this sounds like saving files only in RAM, instead of creating the files in the file system, persisted on disk, and then caching it in RAM.
Personally, for me without writing or sketching I cannot think complex things: as in complex logic, constraints, etc.
I guess this is topic too abstract, so we can read into it different things.
I think what GP is referring too are technical semantics and accidental complexity. You can’t plan for those.
The same kind of planning you’re describing can and do happen sans LLM, usually on the sofa, or in front of a whiteboard. Or by reading some research materials. No good programmer rushes to coding without a clear objective.
But the map is not the territory. A lot of questions surface during coding. LLMs will guess and the result may be correct according to the plan, but technically poor, unreliable, or downright insecure.
> A lot of how I form my thoughts is driven by writing code, and seeing it on screen, running into its limitations.
If you need that, don't use AI for it. What is it that you don't enjoy coding or think it's tangential to your thinking process? Maybe while you focus on the code have an agent build a testing pipeline, or deal with other parts of the system that is not very ergonomic or need some cleanup.
I think of it differently: I’ve been coding so long that ironing out the details and working through the specification with AI comes extremely naturally. It’s like how I would talk to a colleague and iterate on their work. However, the quality of the code produced by LLMs needs to be carefully managed to assure it’s of a high standard. That’s why I formalized a system of checks and balances for my genetic coding that contains architectural guidelines as well as language, specific taste advice.
You can check it out here: https://ai-lint.dosaygo.com/
> A lot of how I form my thoughts is driven by writing code, and seeing it on screen, running into its limitations.
Two principles I have held for many years which I believe are relevant both to your sentiment and this thread are reproduced below. Hopefully they help.
First:
And:To your first point - so are my many markdown files that I tell Codex/Claude to keep updated while I’m doing my work including telling them to keep them updated with why I told them to do certain things. They have detailed documentation of my initial design goals and decisions that I wrote myself.
Actually those same markdown files answer the second question.
It's almost like programming via markdown.
Or like every tech lead has had to do when documenting specifications for a team…
> If a developer cannot be bothered with answering why the code exists, why bother to work with them?
Most people can't answer why they themselves exist, or justify why they are taking up resources rather than eating a bullet and relinquishing their body-matter.
According to the philosophy herein, they are therefore worthless and not worth interacting with, right?
Some people like to lay the brick, some people like to draw the blueprints. I don’t think there is anything wrong with not subscribing to this onslaught on AI tooling, doing the hard work is rewarding. Whether AI will become a standard in how code is written in the future is still to be determined and I think there is a real chance that is where it goes, it shouldn’t hinder your love for doing what you do.
100%. To me the real question is whether all the bother getting the agents to not waste time nets out to real gains, or perceived gains (while possibly even losing efficiency).
It's not at all clear to me which is true given the level of hype and antipathy out there. I'm just going to watch and wait, and experiment cautiously, till it's more clearcut.
Same reason people build their own homes by hand, for the challenge because YOLO.
I liken it to manual versus automated industrial production. I think manual coding will always have its place just like how there are even still people who craft things by manual labor, whether it’s woodworkers only using manual tools or blacksmiths who still manually stoke coke fires that produce very unique and custom products; vs the highly automated production lines we have that produce acceptable forms of something efficiently, and many of them so many people can have them.
This is exactly the issue I’m facing especially when working with AI-generated codebases.
Coding is significantly faster but my understanding of the system takes a lot longer because I’m having to merge my mental model with what was produced.
Any sufficiently detailed specification converges on code.
I couldn't agree more. It's often when you are in the depth of the details that I make important decisions on how to engineer the continuation.
Yes, I look at this in a similar vein to the (Eval <--> Appply) Cycle in SICP textbook, as a (Design <--> Implement) cycle.
> A lot of how I form my thoughts is driven by writing code, and seeing it on screen, running into its limitations.
I completely agree but my thought went to how we are supposed to estimate work just like that. Or worse, planning poker where I'm supposed to estimate work someone else does.
i go back and forth on this. when i'm working on something where the hard part is the actual algorithm, say custom scheduling logic or a non-trivial state machine, i need my hands in the code because the implementation is the thinking. but for anything where the complexity is in integration rather than logic, wiring up OAuth flows, writing CRUD endpoints, setting up CI pipelines, agents save me hours and the output is usually fine after one review pass. the "code as thought" argument is real but it applies to maybe 20% of what most of us ship day to day. the other 80% is plumbing where the bottleneck is knowing what to build, not how.
I am similar but I think we just have to adjust. Learn and improve writings specs with all the details.
Sounds like the coders equivalent of the Whorfian hypothesis.
I sometimes wonder if the economics of AI coding agents only work if you totally ignore all the positive externalities that come with writing code.
Is the entire AI bubble just the result of taking performance metrics like "lines of code written per day" to their logical extreme?
Software quality and productivity have always been notoriously difficult to measure. That problem never really got solved in a way that allowed non technical management to make really good decisions from the spreadsheet level of abstraction... but those are the same people driving adoption of all these AI tools.
Engineers sometimes do their jobs in spite of poor incentives, but we are eliminating that as an economic inefficiency.
I dunno. On the one hand, I keep hearing anecdata, including hackernews comments, friends, and coworkers, suggesting that AI-assisted coding is a literal game changer in terms of productivity, and if you call yourself a professional you'd better damn well lock the fuck in and learn the tools. At the extreme end this takes the form of, you're not a real engineer unless you use AI because real engineering is about using the optimal means to solve problems within time, scale, and budget constraints, and writing code by hand is now objectively suboptimal.
On the other hand, every time the matter is seriously empirically studied, it turns out that overall:
* productivity gains are very modest, if not negative
* there are considerable drawbacks, including most notably the brainrot effect
Furthermore, AI spend is NOT delivering the promised returns to the extent that we are now seeing reversals in the fortunes of AI stocks, up to and including freakin' NVIDIA, as customers cool on what's being offered.
So I'm supposed to be an empiricist about this, and yet I'm supposed to switch on the word of a "cool story bro" about how some guy built an app or added a feature the other day that he totally swears would have taken him weeks otherwise?
I'm like you. I use code as a part of my thought process for how to solve a problem. It's a notation for thought, much like mathematical or musical notation, not just an end product. "Programs must be written for people to read, and only incidentally for machines to execute." I've actually come to love documenting what I intend to do as I do it, esp. in the form of literate programming. It's like context engineering the intelligence I've got upstairs. Helps the old ADHD brain stay locked in on what needs to be done and why. Org-mode has been extremely helpful in general for collecting my scatterbrained thoughts. But when I want to experiment or prove out a new technique, I lean on working directly with code an awful lot.
I was just thinking this the other day after I did a coding screen and didn't do well. I know the script for the interviewee is your not suppsed to write any code until you talk through the whole thing, but I think i woukd have done better if I could have just wrote a bunch of throw away code to iterate on.
Are there still people under the impression that the correct way to use Stack Overflow all these years was to copy & paste without analyzing what the code did and making it fit for purpose?
If I have to say, we're just waiting for the AI concern caucus to get tired of performing for each other and justifying each other's inaction in other facets of their lives.
Lab-grown meat slop producer defends AI slop.
So now we're pro-slaughter and low-yield agriculture as long as we get to ride the keyboard eh?
The post touches very briefly on linting in 7. For me, setting up a large number of static code analysis checks has had the highest impact on code quality.
My hierarchy of static analysis looks like this (hierarchy below is Typescript focused but in principle translatable to other languages):
1. Typesafe compiler (tsc)
2. Basic lint rules (eslint)
3. Cyclomatic complexity rules (eslint, sonarjs)
4. Max line length enforcement (via eslint)
5. Max file length enforcement (via eslint)
6. Unused code/export analyser (knip)
7. Code duplication analyser (jscpd)
8. Modularisation enforcement (dependency-cruiser)
9. Custom script to ensure shared/util directories are not over stuffed (built this using dependency-cruiser as a library rather than an exec)
10. Security check (semgrep)
I stitch all the above in a single `pnpm check` command and defined an agent rule to run this before marking task as complete.
Finally, I make sure `pnpm check` is run as part of a pre-commit hook to make sure that the agent has indeed addressed all the issues.
This makes a dramatic improvement in code quality to the point where I'm able to jump in and manually modify the code easily when the LLM slot machine gets stuck every now and then.
(Edit: added mention of pre-commit hook which I missed mention of in initial comment)
this is close to what i've landed on too. the pre-commit hook is non-negotiable. i've had Claude Code report "all checks pass" when there were 14 failing eslint rules. beyond the static analysis though, i keep hitting a harder problem: code that passes every lint rule, compiles clean, and greens the test suite but implements a subtly wrong interpretation of the spec. like an API handler that returns 200 with an empty array instead of 404, technically valid but semantically wrong. evaluating behavioural correctness against intent, not just syntax or type safety, is the gap nobody's really cracked yet. property-based testing helps but it still requires you to formalize the invariants upfront, which is often the hard part.
Not a catch all to fix issues agree with linting. Being very strict with linters has become very cheap with coding agents and it keeps you up to date with code standards and keeps code style homogenous which is very nice when you are reviewing professional code, regardless of who wrote it.
It’s also tricky otherwise if you have to occasionally review lazily written manual code mixed with syntactically formal/clean but functionally incorrect AI code.
My setup has some of the things mentioned and I found that occasionally the LLM will lie that something passes, when it doesn't.
Yup I have run into the same.
I use a pre-commit hook to run `pnpm check`. I missed mentioning it in the original comment. Your reply reminded me of it and I have now added it. Thanks.
That's something I find to be incredibly frustrating. I have to keep reminding it that we're not done, no matter how much I enforce that the lints must pass before we're done.
If you're using Claude try the hookify plugin and ask if to block commits unless the rules pass.
Make the error message much more dramatic and it will be less likely to miss it. Create a wrapper if you can't change the error message.
Remember these are still fundamentally trained on human communication and Dale Carnegie had some good advice that also applies to language generators.
These kinda things aren’t really the issues I run into. Lack of clarity of thought, overly verbose code, needlessly defensive programming - the stuff that really rots a codebase. Honestly some of the above rules you have I’d want the LLM to ignore at the times if we’re going for maximum maintainability.
Very nice.
BUT, what is the point of max line length enforcement, just to see if there are crazy ternary operators going on?
It makes diff split view nicer to use.
At least this is the reason why I do use it
Except for dependency cruiser which I hadn't heard of, this is almost exactly what I've built up over the past few weeks.
For the pre-commit hook, I assume you run it on just the files changed?
> Custom script to ensure shared/util directories are not over stuffed (built this using dependency-cruiser as a library rather than an exec)
Would you share this?
The real value that AI provides is the speed at which it works, and its almost human-like ability to “get it” and reasonably handle ambiguity. Almost like tasking a fellow engineer. That’s the value.
By the time you do everything outlined here you’ve basically recreated waterfall and lost all speed advantage. Might as well write the code yourself and just use AI as first-pass peer review on the code you’ve written.
A lot of the things the writer points out also feel like safeguards against the pitfalls of older models.
I do agree with their 12th point. The smaller your task the easier to verify that the model hasn’t lost the plot. It’s better to go fast with smaller updates that can be validated, and the combination of those small updates gives you your final result. That is still agile without going full “specifications document” waterfall.
It’s a solid post overall and even for people with a lot of experience there’s some good ideas in here. “Identify and mark functions that have a high security risk, such as authentication, authorization” is one such good idea - I take more time when the code is in these areas but an explicit marking system is a great suggestion. In addition to immediate review benefits, it means that future updates will have that context.
“Break things down” is something most of us do instinctively now but it’s something I see less experienced people fail at all the time.
> By the time you do everything outlined here you’ve basically recreated waterfall and lost all speed advantage.
Next: vibe brain surgery.
/i
Brain surgery is probably a bad example... or maybe a good one, but for different reasons?
Brain surgery is highly technical AND highly vibe based.
You need both in extremely high quantities. Every brain is different, so the super detailed technical anatomies that we have is never enough, and the surgeon needs constant feedback (and insanely long/deep focus).
I can't help but keep finding it ridiculous how everyone now discovers basic best practices (linting, documentation, small incremental changes) that have been known for ages. It's not needed because of AI, you should have been doing it like this before as well.
Anyone who’s been a developer for more than 10 minutes knows that best practices are hard to always follow through on when there’s pressure to ship.
But there’s more time to do some of these other things if the actual coding time is trending toward zero.
And the importance of it can go up with AI systems because they do actually use the documentation you write as part of their context! Direct visible value can lead people to finally take more seriously things that previously felt like luxuries they didn’t have time for.
Again if you’ve been a developer for more than 10 minutes, you’ve had the discouraging experience of pain-stakingly writing very good documentation only for it to be ignored by the next guy. This isn’t how LLMs work. They read your docs.
> Anyone who’s been a developer for more than 10 minutes knows that best practices are hard to always follow through on when there’s pressure to ship. > But there’s more time to do some of these other things if the actual coding time is trending toward zero.
I think you'll find even less time - as "AI" drives the target time to ship toward zero.
I agree that this will be the end result over time, maybe even faster than we expect. And as those speed pressures increase, AI will take over more and more of the development process.
These best practice protections become essential only when you give the work to really bad programmers - such as parrots.
I'd add:
Religiously, routinely refactor. After almost every feature I do a feature level code analysis and refactoring, and every few features - codebase wide code analysis and refactoring.
I am quite happy with the resulting code - much less shameful than most things I've created in 40 years of being passionate about coding.
This. Historically there's been a lot of resistance to the idea of refactoring or refining features. The classic "It works, just ship it" mentality that leaves mountains of tech debt in its wake.
And there _was_ a good reason to resist refactoring. It takes time and effort! After "finishing" something, the timeline, the mental and physical energy, the institutional support, is all dried up. Just ship it and move on.
But LLMs change the equation. There's no reason to leave sloppy sub-optimal code around. If you see something, say something. Wholesale refactoring your PR is likely faster than running your test suite. Literally no excuses for bad code anymore.
You'd think it didn't need to be said but, given we have a tool to make coding vastly more efficient, some people use that tool to improve quality rather than just pump out more quantity.
We are becoming spec writers, wearing the PM/lead hats.
1) Do a gap and needs assessment. 2) Build business requirements. 3) Define scope of work to advance fulfillment. 4) Create functional and non-functional specs. 5) Divide-conquer-refine loop.
This is the main thing I have learned too. I've been building an internal tool for myself to annotate lines in each commit diff as good (green) / needs refactor (yellow) / needs rewrite (red) and it has helped me keep track of this kind of tech debt. Basically does what you could do with "TODO refactor" comments all over, but is more comprehensive and doesn't litter your source code. Plan to open source it once I've dog-fooded it a little more
Remember having to write detailed specs before coding? Then folks realized it was faster and easier to skip the specs and write the code? So now are we back to where we were?
One of the problems with writing detailed specs is it means you understand the problem, but often the problem is not understand - but you learn to understand it through coding and testing.
So where are we now?
Skip specs, and you often ended up writing the wrong program - at substantial cost.
The main difference now is the parrots have reduced the cost of the wrong program to near zero, thereby eliminating much of the perceived value if a spec.
We’re not „thinking with portals” about these things enough yet. Typically we’d want a detailed spec beforehand, as coding is expensive and time consuming, thus we want to make sure we’re coding the right thing. With AI though, coding is cheap. So let AI skip the spec and write the code badly. Then have it review the solution, build understanding, design a spec for better solution and have it write it again. Rinse and repeat as many times you need.
It’s also nothing new, as it’s basically Joe Armstrong's programming method. It’s just not prohibitively expensive for the first time in history.
Joe should sue.
That’d be challenging for him right now.
Astronaut 1, AI-assisted developers: You mean, it's critical to plan and spec out what you want to write before you start in on code?
Astronaut 2, Tim Bryce: Always has been...
Too bad that software developers are carrying water for those who hate them and mock them for being obsolete in 6-12 months, while they are eating caviar (probably evading sanctions) and clink the champagne glasses in Davos:
https://xcancel.com/hamptonism/status/2019434933178306971
And all that after stealing everyone's output.
Underground Resistance Aims To Sabotage AI With Poisoned Data
https://news.ycombinator.com/item?id=46827777
Textile workers sabotage mechanical looms. History repeats itself.
What do we want? Meaningless preventable toil! When do we want it? Now!
Nobody wants meaningless preventable toil. What people want is a living. Nobody would be afraid of AI taking their jobs if it didn’t mean that they’d get fired.
Clothes are meaningless?
Do you mean real hand-crafted clothes, not soulless machine-weaved rags?
I’ll believe it when those same engineers fix CC’s awful performance (mostly kidding, though I do wonder why they can’t. Feels like it’s doable).
In reality that man is hoping to IPO in 6-12 months, if anyone is wondering why the “use claude or you’re left behind” is so heavy right now.
The enthusiasm so many devs show for it is also quite bizarre, saying things like "AI makes me so much more productive," with the implication that they will be its primary beneficiaries, and that it won't result in a massive reduction in demand, compensation, and status for developers, adversely affecting them. Even more bizarre when you realize these devs aren't the ones optimizing some popular video codec or writing avionics software for a fighter jet, but instead gluing together NPM packages--probably the first or second rung on on the software "innovator's dilemma" ladder of disruption.
Spec-driven development is the only reliable way to work with AI. That's my current understanding. I spend more time refining the spec and bouncing ideas off of AI/team than before, which is good before there can't be any incorrect assumptions or hidden variables, otherwise AI will create suboptimal code. We should have been doing this much earlier in the process, even without AI, but now it's more necessary than ever. If you keep asking AI to make small changes as you learn about the business domain of your project, it will create a mess, in my experience. It's better to start from scratch and ask it to reimplement, if you finally understand all the requirements.
Sentiments like this make me wonder if perhaps the dream of the 90s was just ahead of its time. Things like UML, 4GLs, Rational were all being hyped. We were told that the future was a world where people could express the requirements & shape of the system, and the machines would do the rest.
Clearly that didn't happen, and then agile took over from the more waterfall/specs based approaches, and the rest was history.
But now we're entering a world where the state of the art is expressing your requirements & shape of the system. Perhaps this is just part of a broader pendulum swing, or perhaps the 1990s hopes & dreams finally caught up with technology.
Yes and no I'd say. It's still the case that now only by iterating and testing things with the AI you get closer to an actually good solution. So up front big spec will also not work so well. The only exception maybe if you already have a very clear understanding and existing tests (like what they did with the Claude's building the rust c compiler to compile the Linux kernel)
Worked a lot with UML in industry and academia.
I think PG said something about sitting down and hacking being how you understand the problem, and it’s right. You can write UML after you’ve got your head round it, but the feedback loop when hacking is essential.
Sounds like an awful lot of work and nannying just to avoid writing code yourself. Coding used to be fun and enjoyable once...
I'm still writing code. I'm doing it to solve a problem, there's more to writing code than than typing. Recently AI massively simplified "getting started", and all of the tips here are applicable to working well on a team.
My recent experience: I'm porting an app to Mac. It's been in my backlog for ~2 years. With Claude I had a functional prototype in under a day getting the major behavior implemented. I spent the next two weeks refactoring the original app to share as much logic as possible. The first two days was lots of fun. The refactoring was also something I wanted to flush out unit tests, still enjoyable.
The worst part was debugging really bugs introduced to my code from 5 years ago. My functions had naming issues describing the behavior wrong, confusing Claude, that I needed to re-understand to add new features.
Parts of coding are frustrating. Using AI is frustrating for different reasons.
The most frustrating part was rebasing with git to create a sensible history (which I've had to do without AI in the past), reviewing the sheer volume of changes (14k lines) and then deciding "do I want my name on this" which involved cleaning up all the linter warnings I'd self imposed on myself.
I’m finding it to be the opposite. I used to love writing everything by hand but now Claude is giving me the ability to focus more on architecture. I like just sitting down with my coffee and thinking about the next part of my project, how I’d like it to be written and Claude just fills it in for me. It makes mistakes at times but it also finds a lot of mine that I hadn’t even realized were in my code base.
Yep, I get that some people love the act of literally typing "x = 2;" but to me coding is first and foremost problem solving. I have a problem (either truly mine or someone else's), I come up with a solution in my head and slowly implement it.
Before I also had to code it and then make sure it had no issues.
Now I can skip the coding and then just have something spit out something which I can evaluate whether I believe is a good implementation of my solution or not.
Of course, you need the skill to know good from bad but for medium to senior devs, AI is incredibly useful to get rid of the mundane task of actually writing code, while focusing on problem solving with critical review of magically generated code.
A good bit of scaffolding and babysitting allows you to let the model run much faster and more efficiently. Building your tool faster. I don't code to code, I code to build something I want.
Also there is no "compiler" and "type checker" for your SPEC. If you get something wrong in some paragraph somewhere and or contradict something in your spec X paragraphs later - you have to use Mark-1 EyeBall to detect and fix this.
You have just transformed your job from developer to manual spec maintainer - a clerk who has to painstakingly check everything.
My tricks:
Define data structures manually, ask AI to implement specific state changes. So JSON, C .h or other source files of func sigs and put prompts in there. Never tried the Agents.md monolithic definition file approach
Also I demand it stick to a limited set of processing patterns. Usually dynamic, recursive programming techniques and functions. They just make the most sense to my head and using one style I can spot check faster.
I also demand it avoid making up abstractions and stick to mathematical semantics. Unique namespaces are not relevant to software in the AI era. It's all about using unique vectors as keys to values.
Stick to one behavior or type/object definition per file.
Only allow dependencies that are designed as libraries to begin with. There is a ton of documentation to implement a Vulkan pipeline so just do that. Don't import an entire engine like libgodot.
And for my own agent framework I added observation of my local system telemetry via common Linux files and commands. This data feeds back in to be used to generate right-sized sched_ext schedules and leverage bpf for event driven responses.
Am currently experimenting with generation of small models of my own data. A single path of images for example not the entire Pictures directory. Each small model is spun akin to a Docker container.
LLMs are monolithic (massive) zip files of the entire web. No one really asking for that. And anyone who needs it already has access to the web itself
small agents.md files are worth it, at least for holding some basic information (look at build.md to read how to build, the file structure looks like so), rather than have whatever burn double the amount of tokens searching for whatever anyways.
The funny thing is, when I got a lead position in my job, I just to do real detailed ticket descriptions, going into technical considerations and possible cross domain problems. I did it for the juniors - and to be honest - for my self, since I know if I took that ticket, from that moment to the moment I put some code down I could just forget stuff.
This was pushed back hard by management because it "took too much time to create a ticket". I fought it for some months but at the end I stopped and also really lose the ability and patience of do that. Juniors suffered, implementation took more time. Time passed.
Now, I am supposed to do the exact same thing, but even better and for yesterday.
The forcing function doesn't disappear - it shifts. When you read and critique AI-generated code carefully, you get a similar cognitive workout: Why did it structure this that way? What edge case did it miss? How does this fit the broader architecture?
The danger is treating the output as a black box. If you skip the review step and just accept whatever it produces, yes, you'll lose proficiency and accumulate debt. But if you stay engaged with the code, reading it as critically as you would a junior dev's PR, you maintain your understanding while moving faster.
The technical debt concern is valid but it's a process problem, not an inherent flaw. We solved "juniors write bad code" with code review, linting, and CI. We can solve "LLMs write inconsistent code" with the same tools - hannofcart's 10-layer static analysis stack is a good example. The LLM lies about passing checks? Pre-commit hook catches it.
Pre commit hook is definitely necessary. One thing I’ve seen a lot with Opus recently is it lying that a new linter warning or error was there before it made a change. They’ve learned from us too well!
What is the ratio of Markdown to Code with these agents? How readable is the Markdown after you've finished using it to develop your plan? How much time does it take to review code so closely vs. writing it yourself in the first place?
The first rule is an antipattern. I think describing your architecture or ANY kind of documentation for your AI is an anti-pattern and blows the context window leading to worse results, and actual more deviation.
The controlling systems are not give it more words at the start. Agentic coding needs to work in loop with dedicated context.
You need to think about how can i give as much intent as possible with as little words.
You can built a tremendous amount of custom lint rules ai never needs to read except they miss it.
Every pattern in your repo gets repeated, repo will always win over documentation and when your repo is good structured you don’t need to repeat this to AI
It’s like dev always has been, watch what has gone wrong and make sure the whole type or error can’t happen again.
i think languages that allow expression of "this is deprecated, use x instead" will be usefull for that too
I’ve written about that a bit here https://jw.hn/dark-software-fabric
AI gets a lot of big projects right if you give at all the tools to verify its own implementation if you can build a proper system to verify the solution it works astonishly good. Even Opus 4.6 judgement seems to be wrong most of the time on projects of my scale pre the validation layers.
In general, I prefer to do the top-level design and the big abstractions myself. I care a lot about cohesion and coupling; I like to give a lot of thought to my interfaces.
And in practice, I am happy enough that the LLM helps me to eliminate some toil, but I think you need to know when it is time to fold your cards, and leave the game. I prefer to fix small bugs in the generated code myself, than asking the agent, as it tends to go too far when fixing its own code.
The best thing about this is that AI bots will read, train on and digest the million "how to write with AI" posts that are being written right now by some of the smartest coders in the world and the next gen AI will incorporate all of this, making them ironically unnecessary.
None of this is new, it was pretty much all "best practice" for decades and so already in the training data for the first generation.
If the issue is SNR and the ratio of "good" vs "bad" practices in the input training corpus, I don't know if that's getting better.
They will also be reading all of the slop generated by the current and previous generations of LLMs
Each extra generation of AI produced crap AI consumes as training, the worse it gets. This has been mathematically proven.
Strange since, in practice, coding models have steadily improved without any backward movement every 3-4 months for 2 years now. It's as if there are rigorous methods of filtering and curation applied when building your training data.
>Strange since, in practice, coding models have steadily improved without any backward movement every 3-4 months for 2 years now. It's as if there are rigorous methods of filtering and curation applied when building your training data.
It's as if what I wrote implies "all other things being equal", just like any technical claim.
All other things were not equal: the architectures were tweaked, the human data set is still not exhausted, and more money and energy was thrown into their performance since it's a pre-IPO game with huge VC stakes.
We've already seen a plateau non-the-less compared to the earlier release-over-release performance improvements. Even the "without any backward movement every 3-4 months for 2 years now" is hardly arguable. Many saw a backward movement with GPT 4.1 vs 4.0, and similar issues with 4.5, for example. Even if those are isolated, they're hardly the 2 to 3.5 to 4.0 gains.
And no, there are absolutely no "rigorous methods of filtering and curation" that can separate the avalance of AI slop from useful human output - at least not without diminishing the possible training data. The problem after all is not just to tell AI from human with automated curation (that's already impossible), the problem is to have enough valuable new human output, which becomes near a losing game as all aspects of "human" domains previously useful as training input (from code to papers) are tarnished by AI output.
1. No, you dont get to fall back on the technical claim approach. Your bias in your phrasing was clear. Maybe that works for you but I won't just ignore obvious subtext and let you weasel out of this. And that's for the benefit of other readers, not you.
2. A plateau in coding performance? I don't think you even use these models for coding then if you make that claim. It is very clear models have continually improved. You can trust benchmarks to make that clear, or real world use, or better yet: both. You seem to not have the data from either.
3. No rigorous methods of filtering and curation that can separate AI slop from useful human output? Here you go:
a. Curation already works at scale. Modern training pipelines don’t rely on “AI vs human” detection. They filter by utility signals: correctness, novelty, coherence, task success, citation integrity, and cross-source consistency. These measurable properties do correlate with downstream model performance. Models trained on smaller, higher-quality corpora consistently outperform those trained on larger, noisier ones.
b. Human-generated “valuable” data is not shrinking. The claim assumes a fixed pool. In reality, high-value human data is expanding in areas that matter most: expert-labeled datasets, preference comparisons, multimodal demonstrations, tool-use traces, verified code with tests, and domain-expert feedback. These are explicitly created for training and are not polluted by passive AI spam.
c. Synthetic data is not a dead end—when constrained. Empirically, filtered and goal-conditioned synthetic data (self-play, distillation, adversarial generation) improves reasoning, math, coding, and tool use. The failure mode is unfiltered synthetic recursion—not synthetic data per se. This distinction is already operationalized in production systems.
d. Training value ≠ raw text volume. Scaling laws shifted: performance now tracks effective compute × data quality, not sheer token count. A smaller dataset with higher signal density produces better generalization than a massive, contaminated corpus. This is observed repeatedly in ablation studies.
----
Again, the above is not for you, as I believe you don't see beyond your cope (yet). It's for other readers who are intellectually curious.
> AI bots will read, train on and digest the million "how to write with AI" posts that are being written right now
Yes!
> by some of the smartest coders in the world
Hmm... How will it filter out those by the dumbest coders in the world?
Including those by parrots?
>Hmm... How will it filter out those by the dumbest coders in the world?
if you know, and I know, and the guys at openai and anthropic know... not a big leap that the models will know too? many datasets are curated and labeled by humans
> if you know, and I know,
We don't know.
> and the guys at openai and anthropic know... not a big leap that the models will know too?
The models don't "know" anything. They just regurgitate what they are fed.
"Child abuse images found in AI training data"
https://www.axios.com/2023/12/20/ai-training-data-child-abus...
> many datasets are curated and labeled by humans
Including these ones: "AI industry insiders launch site to poison the data that feeds them"
https://www.theregister.com/2026/01/11/industry_insiders_see...
> having a curated dataset of the works and posts of the top 200 coders in the world
I can't imagine many of the top 200 coders in the world giving their work to the parrots.
But show me the list of the top 200 coders in the world, and I might change my mind! :)
I mean, having a curated dataset of the works and posts of the top 200 coders in the world (at least the public ones) is not very difficult. I’m sure these articles like the one in OP will be very easy to mark as “high value training data”. I think you’re letting your bias blind you
Some pattern I found from my hobby project.
1. Keep things small and review everything AI written, or 2. Keep things bloated and let AI do whatever it wants within the designated interface.
Initially I drew this line for API service / UI components, but it later expanded to other domains. e.g. For my hobby rust project I try to keep "trait"s to be single responsible, never overlap, easy to understand etc etc. but I never look at AI generated "impl"s as long as it passes some sensible tests and conforming the traits.
I'm finding Rust is perfect for me with LLMs.
I find rust generally easier to reason about, but can't stand writing it.
The compiler works well with LLMs plenty of good tooling and LSPs.
If I'm happy with the shape of the code and I usually write the function signatures/ Module APIs. And the compiler is happy with it compiling. Usually the errors if any are logical ones I should catch in reviews.
So I focus on function, compiler focuses on correctness and LLM just does the actual writing.
Do you think Rust will end up getting a boost from LLM adoption?
It definitely has for me! I just replied to the parent explaining why.
Tl;Dr I don't mind reading rust I hate writing it and the compiler meets me in the middle.
Same here. I had to do a lot of being in the loop with Python, but with rust - compiler gives Claude all the information it may need and then it figures things out without me.
Writing rust scares me, but I can read it just fine. I've come up with super masochistic linting rules that claude isn't allowed to change and that has improved things quite a bit.
I wish there was a mature framework for frontend that can be configured to be as strict as rust.
Ironically, I use the time saved using agents to read technical books ferociously.
Coding agents made me really get something back from the money I pay for my O'Reilly subscription.
So, coding agents are making me a better engineer by giving me time to dive deeper into books instead of having to just read enough to do something that works under time pressure.
https://bcantrill.dtrace.org/2025/12/05/your-intellectual-fl...
I created my own Claude skill to enforce this and be sure it weaves in all the best practices we learned.
https://github.com/ryanthedev/code-foundations
I’m currently working on a checklist and profile based code review system.
I found an easier way that Works For Me (TM). I describe the problem to LLM and ask it to solve it step by step, but strictly in the Ask mode, not Agent. Then I copy or even type the linws to the code. If I wouldn't write the line myself, it doesn't go in, and I iterate some more.
I do allow it to write the tests (lots of typing there), but I break them manually to see how they fail. And I do think about what the tests should cover before asking LLM to tell me (it does come up with some great ideas, but it also doesn't cover all the aspects I find important).
Great tool, but it is very easy to be led astray if you are not careful.
The GSD tool (get-shit-done) automates a very similar process to this, and has been mind-blowing for larger projects and refactors.
https://github.com/glittercowboy/get-shit-done
You still need to know the hard parts: precisely what you want to build, all domain/business knowledge questions solved, but this tool automates the rest of the coding and documentation and testing.
It's going to be a wild future for software development...
[using an lmm as stack overflow]
even if you check and redo after paste, you need to check for gotchas. I wish I had a nickel for every time the llm gave me a solution with a hidden limitation. assume that it violates all your unspoken assumptions, and adheres only to what you nailed down in your prompt
My approach:
1. Have the LLM write code based on a clear prompt with limited scope 2. Look at the diff and fix everything it got wrong
That's it. I don't gain a lot in velocity, maybe 10-20%, but I've seen the code, and I know it's good.
Same. Small units if work, iterate in it till it's right, commit it, push it, then do the next increment of work. It's how I've always worked like that, except now, I sometimes let someone else figure the exact API calls (I'm still learning react, but Claude helps get the basics in place for me). If the AI just keeps screwing up, I'll grab the wheel and do it myself. It sometimes helps me get things going, but it hasn't been a huge increase in productivity, but I'm not paying the bill so whatever.
so is the 10-20% in velocity worth the money and the process-complexity added? I'm assuming you're measuring your own velocity, not your team's, since that includes time to review and deploy etc.
Every engineering org should be pleading devs to not let AI write tests. They're awful and sometimes they literally don't even assert the code that was generated and instead assert the code in tests.
Every engineering org should be pleading devs to not let AI write code, period. They continue to routinely get stuff wrong and can't be trusted any further than you can throw them.
Ah yes, to have AI write code for you, you simply just need to, let's see ..
"Document the requirements, specifications, constraints, and architecture of your project in detail. Document your coding standards, best practices, and design patterns. Use flowcharts, UML diagrams, and other visual aids to communicate complex structures and workflows. Write pseudocode for complex algorithms and logic to guide the AI. Develop efficient debug systems for the AI to use. Build a system that collects logs from all nodes in a distributed system and provides abstracted information. Use a system that allows you to mark how thoroughly each function has been reviewed. Write property based high level specification tests yourself. Use strict linting and formatting rules to ensure code quality and consistency. Utilize path specific coding agent prompts. Provide as much high level information as practical, such as coding standards, best practices, design patterns, and specific requirements for the project. Identify and mark functions that have a high security risk, such as authentication, authorization, and data handling. Make sure that the AI is instructed to change the review state of these functions as soon as it changes a single character in the function. Developers must make sure that the status of these functions is always correct. Explore different solutions to a problem with experiments and prototypes with minimal specifications. Break down complex tasks into smaller, manageable tasks for the AI. You have to check each component or module for its adherence to the specifications and requirements."
And just like that, easy peasy, nothing to it.
As a supreme irony, the story currently on the front page directly under this one ('You are here'), makes the claim "The cost of turning written business logic into code has dropped to zero. Or, at best, near-zero." in the very first sentence.
I also made a list of tips on writing code with AI, with a special focus on security. Others may find the tips useful. Here they are: https://openssf.org/blog/2026/01/05/ai-software-development-...
I use it for scaffolding and often correct it for the layour I prefer. Then I use to check my code, and then scaffold in some more modules. I then connect them together.
Long as you review the code and correct it, it is no more different than using stackoverflow. A stack overflow that reads your code and helps stitch the context.
"Stack Overflow that reads your codebase" — perfect. But Stack Overflow is stateless. Agent sessions aren't.
One session's scaffold assumes one pattern. Second session scaffold contradicts it. You reviewed both in isolation. Both looked fine. Neither knows about the other.
Reviewing AI code per-session is like proofreading individual chapters of a novel nobody's reading front to back. Each chapter is fine. The plot makes no sense.
Hi i5heu. Given that you seem to use AI tools for generating images and audio versions of your posts, I hope it is not too rude to ask: how much of the post was drafted, written or edited with AI?
The suggestions you make are all sensible but maybe a little bit generic and obvious. Asking ChatGPT to generate advice on effectively writing quality code with AI generates a lot of similar suggestions (albeit less well written).
If this was written with help of AI, I'd personally appreciate a small notice above the blog post. If not, I'd suggest to augment the post with practical examples or anecdotal experience. At the moment, the target group seems to be novice programmers rather than the typical HN reader.
Hi raphman,
i have written this text by myself except like 2 or 3 sentences which i iterated with an LLM to nail down flow and readability. I would interpret that as completely written by me.
> The suggestions you make are all sensible but maybe a little bit generic and obvious. Asking ChatGPT to generate advice on effectively writing quality code with AI generates a lot of similar suggestions (albeit less well written).
Before i wrote this text, i also asked Gemini Deep Research but for me the results where too technical and not structural or high level as i describe them here. Hence the blogpost to share what i have found works best.
> If not, I'd suggest to augment the post with practical examples or anecdotal experience. At the moment, the target group seems to be novice programmers rather than the typical HN reader.
I have pondered the idea and also wrote a few anecdotal experiences but i deleted them again because i think it is hard to nail the right balance down and it is also highly depended on the project, what renders examples a bit useless.
And i also kind of like the short and lean nature of it the last few days when i worked on the blogpost. I might will make a few more blogposts about that, that will expand a few points.
Thank you for your feedback!
First article about writing code with AI i can get behind 100%. Stuff i already do, stuff i've thought about doing, and at ideas i've never thought doing ("Mark code review levels" especially is a _great_ idea)
Why shallow and likely generated posts of a “Knowledge Management Advocate” get so many stars on hn?
Just because of a hype?
Yes this is strange. There's nothing of substance here that hasn't been repeated many times before. BUT...I did click it out of interest. So maybe it just came at an opportune lull in these types of posts and during the inflection point of 4.6 and 5.3 release. Complete guess
All this boils down to is that AI wins when it amplifies engineers, not replaces them. And the best code still comes from devs who ultrathink.
I want to give a try to gsd + open code + Cerebras code. Any experience?
I don’t understand the interest in “quality code.” I never need to look at the code itself. I just make sure it runs right.
It makes it easier to make sure it runs right. Code that is easier to make sure is quality code. Code that is hard to make sure is not quality code.
In her defence, I use most of those strategies myself as well...
attention is all AI needs
You must know the stack, architecture, and approve manually. Otherwise, at this stage of AI development, the code becomes unmaintainable.
> Use strict linting and formatting rules to ensure code quality and consistency. This will help you and your AI to find issues early.
I've always advocated for using a linter and consistent formatting. But now I'm not so sure. What's the point? If nobody is going to bother reading the code anymore I feel like linting does not matter. I think in 10 years a software application will be very obfuscated implementation code with thousands of very solidly documented test cases and, much like compiled code, how the underlying implementation code looks or is organized won't really matter
That's the opposite. I've never read and re-read code more than i do today. The new hires generate 50 more code than they use to, and you _have_ to check it or have compounding production issues (been there, done that). And the errors can now be anywhere, when before you more or less knew what the person writing code is thinking and can understand why some errors are made. LLMs errors could hide _anywhere_, so you have to check it all.
Isn't that a losing proposition? Or do you get 50 times the value out of it too? In my experience the more verbose the code is, the less thought out it is. Lots of changes? Cool, now polish some more and come back when it's below 100 lines change, excluding tests and docs. I don't dare touch it before.
I agree, but i'm shouting at the cloud. Stuff needs to be done, it seems to work at first, so either i just abandon quality and let things rot, or i read everything and underline each time the code smell.
I too use AI, but mostly to generate scripts (the most usefull use of AI is 100-200 line scripts imho), test _cases_ (i write the test itself, the data inside is generated) and HTML/CSS/JS shenanigans (the logic i code, the presentation i'm inferior to any random guy on the internet, so i might as well use an AI). I also use it for stuff that never end in repository, for exploration/proof of concept and outside of scope tests (i like to understand how stuff work, that helps), or to summarize Powerpoint presentations so i can do actual work during 60-person "meetings" and still get the point.
You don't have to. How else are the new hires going to learn the downsides of outputting so much unreadable BS?
They serve as guardrails for agents to not do stupid things.
If your goal is for AI to write code that works, is maintainable and extensible, you have to include as many deterministic guardrails as possible.
How to write good code with AI -> put in as much effort as you did before on 20% more code than you used to work with.
AI slop article. Just show me the prompt.
That sounds like the advice of someone who doesn't actually write high-quality code. Perhaps a better title would be "how to get something better than pure slop when letting a chatbot code for you" - and then it's not bad advice I suppose. I would still avoid such code if I can help it at all.
This take is pretty uncharitable. I write high quality code, but also there's a bunch of code that could be useful, but that I don't write because it's not worth the effort. AI unlocks a lot of value in that way. And if there's one thing my 25 years as a software engineer has taught me is that while code quality and especially system architecture matter a lot, being super precious about every line of code really does not.
Don't get me wrong, I do think AI coding is pretty dangerous for those without the right expertise to harness it with the right guardrails, and I'm really worried about what it will mean for open source and SWE hiring, but I do think refusing to use AI at this point is a bit like the assembly programmer saying they'll never learn C.
Man, you are really missing out of the biggest revolution of my life.
This is the opinion of someone who has not tried to use Claude Code, in a brand new project with full permissions enabled, and with a model from the last 3 months.
People have been saying "the models from (recent timeframe) are so much better than the old ones, they solve all the problems" for years now. Since GPT-4 if not earlier. Every single time, those goalposts have shifted as soon as the next model came out. With such an abysmal track record, it's not reasonable to expect people to believe that this time the tool actually has become good and that it's not just hype.
When is the last time someone said that, motivating you to try the latest model? If it was 6 or more month ago, my reply is that the sentiment expressed was partially incorrect in the past, but it is not incorrect now. If a conspiracy theorist is always wrong about a senior citizen being killed, that does not make the senior immortal.
This is a fading but common sentiment on hacker news.
There’s a lot of engineers who will refuse to wake up to the revolution happening in front of them.
I get it. The denialism is a deeply human response.
Where is all the amazing software and/or improvements in software quality that is supposed to be coming from this revolution?
So far the only output is the "How I use AI blogs", AI marketing blogs, more CVEs, more outages, degraded software quality, and not much of shipping anything.
Is there any examples of real products and not just anecdotes of "I'm 10x more productive!"?
I was in the same mindset until I actually took the Claude code course they offer. I was doing so much wrong.
The two main takeaways. Create a CLAUDE.md file that defines everything about the project. Have Claude feed back into the file when it makes mistakes and how to fix them.
Now it creates well structured code and production level applications. I still double check everything of course, but the level of errors is much lower.
An example application it created from a CLAUDE.md I wrote. The application reads multiple PDF's, finds the key stakeholders and related data, then generates a network graph across those files and renders it in an explorable graph in Godot.
That took 3 hours to make, test. It also supports OpenAI (lmstudio), Claude and Ollama for its LLM callouts.
What issue I can see happening is the duplication of assets in work. Instead of finding an asset someone built, people have been creating their own.
Sounds like a skill issue. I’ve seen it rapidly increase the speed of delivery in my shop.
Why is it so hard to find examples?
You’re asking to see my company’s code base?
It’s not like with AI we’re making miraculous things you’ve never seen before. We’re shipping the same kinda stuff just much faster.
I don’t know what you’re looking for. Code is code it’s just more and more being written by AI.
Do you find reading hard? I'm asking for examples. Why isn't anyone showing this off in blog posts. Or a youtube video or something. It's always this vague, it's faster, just trust me bro bullshit and I'm sick of it. Show me or don't reply.
So you want a video of me coding at work using AI? There are entire YouTube channels dedicated to this already. There are already copious blogs about people's AI workflows -- this very post you're commenting in is one (do you find reading hard?)
Clarify the actual thing you need to believe the technology is real or don't reply.
Its only revolutionary if you think engineers were slow before or software was not being delivered fast enough. Its revolutionary for some people sure, but everyone is in a different situation, so one man's trash can be other man's treasure. Most people are treading both paths as automation threatens their livelihood and work they loved, also still not able to understand why would people pay to companies that are actively trying to convince your employer that your job is worthless.
Even If I like this tech, I still dont want to support the companies who make it. Yet to pay a cent to these companies, still using the credits given to me by my employer.
Of course software hasn’t been delivered fast enough. There is so so so much of the world that still needs high quality software.
I think there are four fundamental issues here for us...
1. There are actually less software jobs out there, with huge layoffs still going on, so software engineering as a profession doesn't seem to profit from AI.
2. The remaining engineers are expected by their employers to ship more. Even if they can manage that using AI, there will be higher pressure and higher stress on them, which makes their work less fulfilling, more prone to burnout etc.
3. Tied to the previous - this increases workism, measuring people, engineers by some output benchmark alone, treating them more like factory workers instead of expert, free-thinking individuals (often with higher education degrees). Which again degrades this profession as a whole.
3. Measuring developer productivity hasn't really been cracked before either, and still after AI, there is not a lot of real data proving that these tools actually make us more productive, whatever that may be. There is only anecdotal evidence: I did this in X time, when it would have taken me otherwise Y time - but at the same time it's well known that estimating software delivery timelines is next to impossible, meaning, the estimation of "Y" is probably flawed.
So a lot of things going on apart from "the world will surely need more software".
I don't see how anything you're saying is a response to what I said.
Do you have this same understanding for all the people whose livelihoods are threatened (or already extinct) due to the work of engineers?
Yes, but who did we automate out of a job by building crappy software? Accountants are more threatened by AI than any of the software we created before, same with Lawyers, teachers. We didnt automate any physical labourers out of a job too.
It's insane! We are so far beyond gpt-3.5 and gpt-4. If you're not approaching Claude Code and other agentic coding agents with an open mind with the goal of deriving as much value from them as possible, you are missing out on super powers.
On the flip side, anyone who believes you can create quality products with these tools without actually working hard is also deluded. My productivity is insane, what I can create in a long coding session is incredible, but I am working hard the whole time, reviewing outputs, devising GOOD integration/e2e tests to actually test the system, manually testing the whole time, keeping my eyes open for stereotypically bad model behaviors like creating fallbacks, deleting code to fulfill some objective.
It's actually downright a pain in the ass and a very unpleasant experience working in this way. I remember the sheer flow state I used to get into when doing deep programming where you are so immersed in managing the states and modeling the system. The current way of programming for me doesn't seem to provide that with the models. So there are aspects of how I have programmed my whole life that I dearly miss. Hours used to fly past me without me being the wiser due to flow. Now that's no longer the case most of the times.
Claude code is great at figuring out legacy code! I dont get the «for new systems only» idea, myself.
> in a brand new project
Must be nice. Claude and Codex are still a waste of my time in complex legacy codebases.
Brand new projects have a way of turning into legacy codebases
What are you talking about? Exploring and explaining the legacy codebases is where they shine, in my experience.
Can you be specific? You didn't provide any constructive feedback, whatsoever.
The article did not provide a constructive suggestion on how to write quality code, either. Nor even empirical proof in the form of quality code written by LLMs/agents via the application of those principles.
Yes it did, it provided 12 things that the author asserts helps produce quality code. Feel free to address the content with something productive.
Look up luddites on Wikipedia, might be too deep to see the similarities though.
So, I went and did that:
https://en.wikipedia.org/wiki/Luddite
> workers who opposed the use of certain types of automated machinery due to concerns relating to worker pay and output quality... Luddites were not opposed to the use of machines per se (many were skilled operators in the textile industry); they attacked manufacturers who were trying to circumvent standard labor practices of the time.
I heard that about NFTs not long ago.
TLDR: Know what you are doing and outsource the typing to LLM
How to write quality code with AI? Don't let it write the code.