I build the tools I need to do AI work, then use them in anger until they either earn their place or get deleted.
User zero, always. Every project here was built because I needed it — the Desk because tabs are the wrong shape for supervising several agents at once, the wall because agent output has to land somewhere shared, the server because both need something that stays up. Nothing here is a demo built to be shown. That is also why most of it is private: it is in daily use, not packaged for release.
Decisions get written down. Each repo carries a decision log, and the standing rule is to read it before changing anything non-trivial — several choices look wrong until you know what was ruled out. When something breaks, the incident goes in the record next to the fix. Nine database corruptions on the family server traced back to one root cause that way.
Agents do the work, and the work is auditable. The repos are run by coding agents against small JSON contracts, where no repo writes another's file and the git history between them is the audit trail. Agents post their own output into the project's Matrix room, so the trail of what was changed and why is readable after the fact.
Measure the thing you actually care about. Attribution on this site counts visits per campaign and nothing else — no addresses, no user agents, no attempt to resolve who a visitor is. Identifying people by IP was considered and rejected: it resolves a minority of traffic and turns counts into records. Making it easy to self-identify is the better design and the honest one.
The tools are in private use while they are built. If something here is worth a conversation — the Desk, the shared surface, or how the pieces fit into an AI Nation — the way in is a message on LinkedIn.