How I use skills for everything

Andy Maguire at PostHog keeps his AI agents' project memory in versioned skill files that the agents update after every session and teammates can read.

PostHog makes product analytics software and runs a shared store of agent skills. A skill is a bundle of instructions and files that an AI agent loads when a task needs it. Andy Maguire builds PostHog's self-driving product, which uses AI agents to find problems and propose fixes. He describes how 39 skills hold his project trackers, runbooks, daily review lists and handover notes, and how his agents keep them current.

• Each project skill holds a backlog, a change log covering 14 days and a handover file. At the end of every session the agent records new issues, pull requests and where the work stopped.
• A small table tells the agent where a note belongs, based on whether it will still be true in 30 days and who keeps it current.
• A debugging runbook grows with every investigation, because the agent writes up each new class of problem so the next session does not start from zero.
• Before a holiday, Maguire handed over his work in one team channel message. Almost every point in it linked to a skill that a teammate or their agent could load.

This is a PostHog engineer describing PostHog's own skills store, so the account favours the product. A team without that store can still keep an agent-maintained handover file with written rules for what goes where, which is an inference and is not the author's claim.

Derived by Working Surface from the article; more in the Playbook. Source line: I don't rely on a model's hidden memory feature that I can't read.

21 October 2026: as AI tools multiply code, documents and prototypes, the work that counts moves to judging, reviewing and keeping track of what they produce.