AGENT ATLASa field guide to agent workflows
Back to Multi-agent
Multi-agent

Blackboard

Agents collaborate by reading and writing a shared workspace.

N agentsAdvanced

How it works

  1. 1Maintain one shared structured workspace — the blackboard.
  2. 2Each agent reads the current state, contributes its piece, writes back.
  3. 3A controller decides who acts next based on the board's state.
  4. 4Continue until the board reaches a done condition.

Use it when

Loosely-coupled agents that contribute to a common evolving artifact — a shared plan, document, or state — without rigid ordering.

Reach for something else when

Simple linear tasks. Shared mutable state needs care to avoid conflicting writes.

Where you stay in the loop

Agents co-write a shared artifact; you set the done-condition and arbitrate conflicting writes. Shared mutable state rewards a clear owner watching for drift.

In the wild

Several agents incrementally fill a shared project plan — one adds tasks, another estimates, another flags risks.

Hand this to your agent

Use a blackboard pattern.

Keep a single shared document (I'll paste its current state each turn).
Define 3 contributor roles.

On each turn, pick the role that can most improve the current state, have it
read the board, make its contribution, and output the full updated board.
Stop when these done-conditions are met: <conditions>.

Initial board: <...>

Replace the <…> placeholders, paste it into your agent, and it'll scaffold the workflow with you.