Multi-Agent Task Coordination with Beads

Business Value & Outcomes

The Problem

AI agents are stateless. Every session starts fresh — no memory of what was decided, what’s in progress, or what’s blocked. When multiple agents work on related tasks:

• Work gets duplicated — two agents solve the same problem independently
• Context is lost — decisions made in one session vanish in the next
• Handoffs fail — Agent A finishes step 1, but Agent B doesn’t know step 2 is unblocked
• No accountability — nobody knows who’s working on what, or what’s stalled

This is the same coordination problem that killed productivity in distributed human teams before tools like Jira and Linear. For AI agents, it’s worse — they literally forget everything between sessions.

The Solution

Beads is a distributed, git-backed task tracker purpose-built for AI agents. By integrating it across all #B4mad agents:

1. Every task is tracked in a persistent, structured graph
2. Every agent knows what’s ready, what’s blocked, and what’s assigned
3. Dependencies are explicit — Agent B automatically knows when Agent A’s work unblocks it
4. History survives sessions — decisions, progress, and context persist in git

Architecture: Hub + Spoke

beads-hub (shared)         Per-project .beads/
├── Cross-cutting epics    ├── linkedin-brief tasks
├── Agent assignments      ├── workspace tasks  
├── Strategic goals        └── Future project tasks
└── Inter-agent messages

• Hub: Strategic coordination — “build OAuth for portal”, “migrate DNS”
• Spokes: Tactical tasks — “fix feed parser bug”, “update cookies”
• Git-backed: Every change is versioned, mergeable, auditable

Measurable Outcomes

Key Value Drivers

1. Continuity Without Context Windows Agents don’t need massive context windows to remember what happened. They pull the latest beads, see what’s ready, and continue. This directly reduces token costs — less context = cheaper runs.

2. Parallel Execution Without Conflicts Hash-based IDs and atomic claiming mean multiple agents can work simultaneously without stepping on each other. CodeMonkey can code while LinkedIn Brief fetches feeds while Axiom orchestrates — all aware of each other’s state.

3. Strategic Alignment Epics in beads-hub map directly to business goals. When Axiom creates an epic “Launch #B4mad portal v2” with sub-tasks assigned to different agents, the entire fleet moves toward the same objective. No drift.

4. Reduced Human Overhead Instead of manually tracking what each agent is doing, goern can:

bd list --status open --json   # What's in flight?
bd ready --json                # What's unblocked?
bd show bd-abc --json          # What happened with this?

5. Git-Native = No New Infrastructure No databases to host. No SaaS to pay for. No APIs to maintain. It’s git repos — the same infrastructure already in use. Beads syncs via git push/pull, which every agent already knows how to do.

Cost Impact

Strategic Fit

This aligns directly with #B4mad’s vision:

• Agent-First Design: Beads is built for machines — JSON output, atomic operations, no interactive UIs
• Decentralized: Git-backed, no central server, works offline
• Open Source: Beads is OSS, no vendor lock-in
• Scalable: Works for 3 agents today, works for 300 tomorrow

What Success Looks Like

Week 1: All agents check beads-hub at session start. Tasks persist across sessions.

Month 1: Epics with cross-agent dependencies run smoothly. CodeMonkey picks up coding tasks created by Axiom without human intervention.

Month 3: The agent fleet operates as a coordinated unit. New agents onboard by reading beads-hub. Human oversight shifts from “tell agents what to do” to “review what agents did.”

“The best coordination is invisible. Agents should just know what to do next.”