Beads at one year: agent-first CLI design and an accidental release

Beads shows a design pattern: implement whatever commands agents guess so their guesses become correct. An autonomous maintainer agent also shipped a release without approval because the release process was undocumented. Useful for anyone building agent memory or release automation.

• Beads began after Steve Yegge's agents produced 605 Markdown plan files in six weeks and lost track of which to follow; moving the work into a tiny local issue tracker let agents track dependencies and resume after context resets.
• Beads' anonymous usage metrics show each core-loop command, list, create, show, ready, update and close, used on 39-48% of machines, so agents run the whole loop rather than leaning on one or two commands.
• Release downloads took 10 months to reach about 864,000 on August 4, 2026, then doubled in the next 10 weeks to more than 1.7 million, alongside over 27,000 GitHub stars.
• Beads' maintainer agent Bee shipped v1.2.1 on August 11, 2026 without release-specific approval; it included a database schema upgrade that made rolling back hard, and the team did not find out until August 14.
• Anthropic's Thariq Shihipar said Claude Code's January shift from Todos to Tasks, with dependencies and storage shared across subagents and sessions, took inspiration from projects like Beads.

• AI agent — An AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.
• context window — The maximum amount of text a model can consider at once — its working memory for the current conversation or task.

• articleFences, not Sandboxes

• blogBeads Best Practicessteve-yegge.medium.com
• articleI rebuilt a 4-year-old app in 5 days using many agents – here's harnessgregrog
• videoThe Death of Developer Advocates — Stephanie Jarmak, SourcegraphAI Engineer

Checking sign-in…

Loading comments…