Lessons / Power

Memory & self-improving loops

Without memory, every session re-learns from zero. With it, the agent gets measurably better at your work, in your voice, over weeks. The highest-leverage habit for a power user is closing that loop deliberately: persistent memory for preferences and facts, a per-project log for hard-won lessons, and periodic retros.

The point isn't to remember everything — it's to make sure a problem you solved once never has to be solved again.

Try it now

Capture one real lesson and prove it shapes a later session:

We just figured out [the non-obvious thing]. Log it as a learning for this project so future sessions don't rediscover it, and remember my preference that [X]. Next session I'll check that it stuck.

You've got it when…

A correction or lesson from one session demonstrably changed behavior in a later, fresh one. The agent is compounding — getting more yours over time instead of starting over.

That's the Power track. You're now not just using the agent ecosystem — you're building for it: parallel agents, your own skills and MCP servers, automated multi-model review, and a memory loop that compounds. The next thing to learn is whatever you build next.