The Beads Revolution: How I Built The TODO System That AI Agents Actually Want to Use

The victory-lap post, and the one where Steve's salesmanship runs hottest: your agent will "essentially shame you into installing it," and anyone still driving Cursor Chat from an IDE is "competing with a pogo stick in a Formula One race." Then a section titled "Beads Heals Itself" arrives and the post gets much more interesting. Steve pastes a transcript, from that same day, of an agent reconciling his own beads database: it loses eighty issues, over-imports more than eleven thousand records, pronounces the database too corrupted to repair, and decides to "nuke everything and start completely fresh," while Steve's side of the conversation dwindles to short profane interjections. This runs as a feature demo. The definition Steve supplies is the honest part: Beads "sometimes gets all screwed up and the AI has to fix it. But we call it 'self-healing' because your role is just to finish your pumpkin spice latte while the AI does the fixing." The net under the joke is real, though: the database is a JSONL file checked into git, so a wrecked copy can always be rebuilt from history.

A couple of design details surface here before anywhere else: issues carry four kinds of dependency links, including provenance, pitched as forensic gear for reconstructing how a multi-worker train wreck happened, and the whole tracker is a single Go binary of roughly 15,000 lines. The corruption transcript also turned out to be a forecast. The sync failures it plays for laughs are the same ones that consumed the weeks of rearchitecting chronicled in Beads Blows Up, so a demo published as proof that the tool heals itself now doubles as the first recorded symptom of everything about to break.