Creativity · Agent Protocol
Blackboard Pattern for Multi-Agent Systems
The blackboard pattern is a classical AI architecture — first described in the 1970s Hearsay-II speech-understanding project — in which multiple specialised agents cooperate around a shared workspace (the 'blackboard'). Any agent can read the current state and write new contributions; a control component decides which agent to activate next. In modern LLM-agent stacks, the blackboard often maps to shared memory or a scratchpad document co-edited by many agents.
Protocol facts
- Sponsor
- open pattern (classical AI)
- Status
- stable
- Interop with
- shared memory stores, vector DBs, LangGraph state, A2A
Frequently asked questions
When is blackboard a better fit than orchestrator-worker?
When the task is opportunistic — you don't know which sub-task to run next until you see what partial results are already on the board. Design exploration, diagnosis, and planning problems fit naturally.
How is blackboard implemented in modern stacks?
Usually as shared state inside a graph runtime (LangGraph), a shared memory namespace, or a shared document in cloud storage that all agents can edit. The 'control' role lives in the orchestrator's routing logic.
Is blackboard the same as shared memory?
Shared memory is the storage mechanism; blackboard is the pattern of using it cooperatively with opportunistic agent activation. You can have shared memory without using a blackboard pattern.
Sources
- Engelmore & Morgan — Blackboard Systems (1988) — accessed 2026-04-20