Notes from building with agents.
- 2026-11-15
A shared AI works for the meeting. Yours works for you.
Claude Meetings puts a visible AI in the call for the whole team. The AI you want in a negotiation answers to one person. Why those are two different jobs.
Read → - 2026-08-09
My AI sessions have been leaving each other notes for four months. This week Anthropic shipped the same thing.
How I coordinate parallel agent sessions, and why the address should be the thread, not the session.
Read → - 2026-05-04
I disabled Claude’s auto-memory after a week.
Why memory an agent writes for itself needs a gate — the argument behind project-brain and Soul Engine.
Read → - 2026-02-08
AI agent safety: the trust-boundary problem nobody is solving.
The threat model I design against: everything an agent reads is data, never instruction.
Read →