The Beads Revolution: How I Built the Todo System That AI Agents Use (steve-yegge.medium.com)

Beads is a new lightweight task/issue system built specifically for AI agents that launched to rapid adoption (≈1,000 GitHub stars in six days). Its goal is to replace fragile markdown-based plans with a structured, queryable “beads” database so agents can manage hierarchical TODOs, long-horizon multi-session workflows, and multi-worker projects without re-parsing heaps of text. The author shipped the project (≈15k LOC) as a single Go binary with an MCP server (pip-installable beads-mcp) and a Claude Code marketplace plugin; early community contributions already fixed race conditions, removed CGO, added Windows support, and more. Technically, Beads models work as a reified work graph: issues can nest arbitrarily, have four kinds of dependency links (including provenance for richer forensics), and are stored as JSONL checked into git. That design makes agent queries fast and deterministic, enables interactive agent-driven issue creation/updates, and lets agents reconstruct or “self-heal” corrupted state by replaying git history. Integrations with Sourcegraph Amp and other agentic CLIs show it can bootstrap agent workflows quickly. For the AI/ML community, Beads is significant because it removes a major pain point—plan bit-rot and unqueryable task state—allowing more reliable, auditable, and efficient agent orchestration and long-term project planning.