Discussion about this post

User's avatar
Pawel Jozefiak's avatar

The auto-memory system (MEMORY.md) is the one update here I'd watch most carefully - it shifts context from something you manually curate to something the agent builds alongside you. Whether that's better depends entirely on how disciplined the capture logic is.

I've been thinking about this from the other direction: what goes in your context files matters as much as having them at all (https://thoughts.jock.pl/p/how-i-structure-claude-md-after-1000-sessions).

The /simplify and /batch commands feel like quality-of-life improvements; the memory layer feels like a different architectural assumption.

No posts

Ready for more?