Discussion about this post

User's avatar
Thomas Landgraf's avatar

Thanks a lot! 🙏 I made the same observation with context priming — and I started naming my prime commands after brilliant colleagues I’ve worked with over the years. It’s fun, and honestly a bit nostalgic, to see Claude Code suddenly take on their skills and quirks. It’s like pair programming with ghosts of engineering past — in a good way. 😄

Your journaling approach sounds brilliant. Love how you’re turning session history into pattern recognition fuel. That’s exactly the kind of muscle memory we’re all trying to build — both for ourselves and our AI partners.

Expand full comment
Mike Hernandez's avatar

This is such a great post! Thanks for sharing your insights. I think everyone is going through an evolution of learning as we use Claude Code. And we discover how much managing context helps reduce which version of non-deterministic Claude we'll get with each session. Will give your learnings a try.

Have been working on a small CLI app using the Claude SDK on a Pro plan, intentionally with the constraints. Smaller iterations and simpler code (no web framework bloat) I figured would give me smaller and faster Claude Code sessions. Started with the context priming by IndyDevDan, pointing specifically to targeted files.

Then I added journaling with every commit and session (checkpoints). It's been useful when running into repeat bugs. I'll point to the journal when a new bug smells familiar, and Claude will recognize the pattern, avoid tunnel vision, and fix it immediately.

That has been the most impressive observation around memory I've had so far.

Hope your deep dive into memory makes its way into a Simon Willison shout out!

Expand full comment

No posts