Hey Kamil, I like caveman! The only complaint I have about it is you'll have to invoke it every time. Another alternative is context-mode plugin, which just manages context from mcps in the background via hooks.
This is the right way to think about it. The real issue is that most teams are using these tools with almost no operating discipline. New chat for every task, same context pasted 40 times, Opus used for throwaway drafting, giant catch-all projects, no shared team standard.
This is exactly why we treat Claude/ChatGPT spend as a per-client line item now, not a generic 'tools' expense. Two habits that helped most: pin the cheapest model that actually works for the task (most extraction does not need Opus), and never resend the same 30-page client file in every prompt — set it up once. Bookkeeping rule: if you would not fax the same document twice, do not paste it twice.
Fantastic - on my top to do list thanks!
Hey Kamil, I like caveman! The only complaint I have about it is you'll have to invoke it every time. Another alternative is context-mode plugin, which just manages context from mcps in the background via hooks.
Great timing on this. Cost management for AI tools is becoming increasingly important.
This is the right way to think about it. The real issue is that most teams are using these tools with almost no operating discipline. New chat for every task, same context pasted 40 times, Opus used for throwaway drafting, giant catch-all projects, no shared team standard.
This is exactly why we treat Claude/ChatGPT spend as a per-client line item now, not a generic 'tools' expense. Two habits that helped most: pin the cheapest model that actually works for the task (most extraction does not need Opus), and never resend the same 30-page client file in every prompt — set it up once. Bookkeeping rule: if you would not fax the same document twice, do not paste it twice.