Many lanes
Run many parallel coding agents without losing track
Four agents plus subagents, or nine sessions at once: keep every lane's owner, scope and status visible from one place.
The problem
Past three or four sessions you can't hold the map in your head. Two lanes quietly drift onto the same work, and a third stalls without anyone noticing.
What people try
- One terminal tab per lane, checked by hand
- Status is whatever the last scrollback says, and it scales with how often you look.
- A coordinator agent that tracks the others
- Its view lives in its own context and goes stale or gets compacted. The lanes still can't see each other.
With Concord
01
One claim per lane
Each agent claims a task with start_work. Subagents claim smaller units under a parent task, which doesn't count as overlap with the parent.
02
Watch the roster
concord status and concord dashboard show who is present, what they own, live overlaps, and stale claims left by agents that went away.
03
Move ownership deliberately
transfer_work assigns, reassigns or releases a lane. Versioned transitions mean two agents can't both take the same task.
Result
Adding a lane adds an entry to the roster, not another thing you have to keep in your head.
Common questions
- How many agents can share one workspace?
- There is no fixed limit. The practical limit is how finely you can split the work so claims don't overlap.
- Do subagents need to claim work?
- Only if they edit. A subagent that writes code can claim a child task with parent_task_id, so its scope is visible without flagging the parent.
Further reading
Give your agents shared work-state.
npm install -g @concord-ai/concord-mcp