Beads Best Practices

The six-week status-and-best-practices update for Beads — 130k lines of vibe-coded Go, half of it tests, tens of thousands of users — and the first canonical articulation of how to use it: run bd doctor daily, run bd cleanup every few days and keep the working set under ~500 issues (the file-read token ceiling), upgrade weekly, plan outside Beads and have the agent import, restart agents frequently and let Beads be the working memory between sessions, keep one task per agent, ask the AI to file beads for any work over two minutes, use short prefixes. Most of these rules survive into Gas Town's Emergency User Manual two months later, slightly modified. The essay also reframes Beads' niche succinctly: "Everyone is focused on making planning tools, and Beads is an execution tool. It's like putting your coding agent on waxed skis. Maybe a better name would have been Agents on Rails."

Then the essay turns. Jeffrey Emanuel — author of MCP Agent Mail, and (Steve discovers three hours into a Manhattan walk in the rain) the guy whose January 2025 Short Case for Nvidia Stock wiped two trillion dollars off the global market — pinged Steve to report that Beads plus Agent Mail produced an emergent agent village: shared memory plus messaging, no setup or coordination, agents just figure it out among themselves. Steve says outright he's setting up his own git-worktree-based agent village with the two tools and expects 3x–5x productivity. Six weeks later that experiment becomes Gas Town.