11 points | by FrancescoMassa 13 hours ago ago
4 comments
How do you handle prompt caching? A lot of cost savings for a single model chat come from cache hits on the conversation context, and switching models invalidates that cache — the new model has to reprocess everything at full input price.
What does this solve that well used subagents doesn't solve already?
On our tests subagents & well used workflows are 20-30% more expensive for context & token efficiency
How do you handle prompt caching? A lot of cost savings for a single model chat come from cache hits on the conversation context, and switching models invalidates that cache — the new model has to reprocess everything at full input price.
What does this solve that well used subagents doesn't solve already?
On our tests subagents & well used workflows are 20-30% more expensive for context & token efficiency