Written from the work, not from a content calendar.
Guides explain a mechanism you can apply whether or not you use Unforgetter. Field notes record what building and testing the product showed, including what went wrong. Each piece is dated, attributed and has a Markdown version for agents.
Built-in AI memory creates better silos
Every major agent now remembers you. That solves the single-vendor problem and makes the cross-agent problem worse. A field note from running three agents on the same work.
Written 27 Aug 2026
Context files vs self-maintaining working truth
CLAUDE.md, AGENTS.md and their cousins are the best cheap tool for agent context. This guide is about the exact point where they stop working, and what has to take over.
Written 27 Aug 2026
Memory is not context: what an agent actually needs to continue work
Storing everything an agent ever saw is not the same as giving the next agent what it needs. A concrete task bundle shows the difference.
Written 27 Aug 2026
What a context compiler does
Include, exclude, scope, freshness and a single write path. The mechanism behind 'the right context for every agent task', explained without hand-waving.
Written 27 Aug 2026
The Unforgetter pilot: what we measure and why
The pilot runs 25 August to 8 September 2026. This note publishes the scorecard, the baseline problem and the go/no-go criteria before any result, so the numbers cannot be chosen afterwards.
Written 27 Aug 2026
Give every agent the same memory.
The alpha is for people who use one or more agents for ongoing work and want the memory to stay current.