“allow it to change its own harness for better context management, etc”
It gets a little too hand-wavey there. If it were that simple, wouldn’t coding agents be doing this already? In my experience the memories CC creates for itself are like any other LLM writing: overly verbose and full of irrelevant details, often missing the point of the feedback that caused it to create the memory in the first place. LLMs are really bad at knowing how to prompt LLMs, and that makes perfect sense for because they’re incapable of introspection.
I agree that current memory systems are pretty bad, and I think that’s because memory is a prompted behavior instead of a learned one. In theory, if memory was an emergent behavior instead of a prompted one, it would be a lot better.
I think you’re right that changing its own harness would be bad and skew towards prompt engineering instead of learned behavior. So maybe instead it could start with a harness with memory CRUD tools and then learn how to most effectively use them.