Blackboard Multi-Agent
Picture a team gathered around a shared whiteboard: anyone who can add something useful steps up, writes it, and steps back. Blackboard multi-agent systems bring that classic pattern to language models. A common workspace holds the evolving state, and at each turn the agent best suited to contribute is chosen based on what is currently on the board. The cycle repeats until the problem is solved, with no fixed script dictating who speaks when.
*These are learning tools, not prompts. They teach you to write your own, think of them as training wheels that guide you while you learn, then fall away once you can ride on your own.
A Shared Board, Chosen Contributors
Many multi-agent systems run on a fixed conversation order: agent A speaks, then B, then C, regardless of whether that is who the problem needs next. That rigidity wastes turns on agents with nothing to add and can miss the agent who does. The blackboard architecture, a long-standing idea in artificial intelligence, offers a more fluid alternative.
All agents share one workspace, the blackboard, that holds the current state of the problem. Instead of a fixed speaking order, a control step looks at what is on the board and selects the agent best positioned to contribute next. That agent reads the shared state, adds its contribution, and the board updates. Then the selection happens again, driven entirely by the evolving content rather than a predetermined script.
This continues, read, select, contribute, until the board holds a solution or the agents reach consensus. Because everyone works against the same shared state and whoever acts is chosen by relevance, the system adapts its collaboration to the problem as it unfolds. It is a natural fit for tasks where you cannot know in advance which expertise will be needed at each step.
The blackboard is a shared workspace every agent can read and write. A control loop picks who acts next from the board's current content, so collaboration follows the problem rather than a fixed turn order. The loop repeats until the board holds a solution.
The Blackboard Process
Read the board, select an agent, contribute, and repeat
Post the Problem to the Board
The task and any starting information are written to the shared blackboard, which becomes the single source of truth all agents read from. Every agent sees the same evolving state.
A diagnostic case is posted: symptoms, history, and initial test results, all visible to the specialist agents.
Select Who Contributes Next
A control step examines the current board and chooses the agent best suited to move the problem forward right now. The choice depends on the state, not a fixed order, so the most relevant expertise is called on.
Given the test results on the board, the control step selects the specialist whose expertise those results most concern.
Read, Contribute, Update
The selected agent reads the full board, adds its contribution, a finding, a hypothesis, a partial result, and writes it back. The shared state advances, and the new content sets up the next selection.
The specialist adds a likely cause and a recommended follow-up test, updating the board for the next contributor.
Repeat Until Solved
The select-and-contribute cycle repeats, each turn driven by the board's latest state, until the problem is solved or the agents reach consensus. The collaboration path is emergent, shaped by the problem rather than a script.
After a few rounds of contributions, the board converges on a diagnosis the agents agree on.
See the Difference
A fixed speaking order versus a shared board
Fixed-Order Agents
Agents speak in a preset order, whether or not they have something useful to add at that moment. The order is decided before the problem is known.
Turns are wasted on agents with nothing to contribute, and the agent the problem actually needs may not be next in line.
Blackboard Multi-Agent
All agents read and write a common workspace, and a control step picks whoever can best contribute given the current state.
The right expertise is called on at the right moment, and collaboration adapts to the problem as the board evolves.
Practice Responsible AI
Always verify AI-generated content before use. AI systems can produce confident but incorrect responses. When using AI professionally, transparent disclosure is both best practice and increasingly a legal requirement.
Most US states are actively legislating AI transparency and accountability. Critical thinking remains your strongest tool against misinformation.
Blackboard Multi-Agent in Action
See how the shared board drives collaboration
A complex case is posted; several specialist agents can read it, but only one is needed at a time.
Round 1: the imaging specialist is selected and adds a finding.
Round 2: that finding shifts the board, so the lab specialist is now the best fit and contributes next.
Result: the sequence of contributors is decided by the evolving case, not fixed in advance.
Build a plan that needs legal, financial, and technical input, but not in a known order.
Each agent contributes when the board calls for it: a legal constraint surfaces, prompting the finance agent to revise its numbers, which in turn raises a technical question.
Result: the plan comes together through whichever expertise the current state requires next.
The agents must agree on a single recommendation.
Contributions accumulate on the board until no agent has an objection or addition to make.
Result: the loop halts at consensus, and the board's final state is the shared recommendation.
When to Use Blackboard Multi-Agent
Best when the needed expertise varies as the problem unfolds
Perfect For
When you cannot know in advance which expertise each step needs, choosing contributors from the board's state fits far better than a fixed order.
Problems where every agent benefits from seeing the same evolving picture suit a common blackboard as the single source of truth.
When you have several specialists and only some are relevant at any moment, relevance-based selection avoids wasting turns.
Tasks where the path to a solution is not known ahead of time benefit from a collaboration that adapts as the board evolves.
Skip It When
If the order of steps is clear and stable, a simple pipeline is easier than a board with dynamic selection.
When one agent can do the whole job, a shared board and selection loop are unnecessary machinery.
Each selection step and shared-state pass costs model calls, so a strict latency budget may favor a leaner setup.
Use Cases
Where Blackboard Multi-Agent delivers the most value
Collaborative Diagnosis
Bring specialist agents together on a shared case, letting each contribute when the evolving evidence calls for its expertise.
Complex Planning
Assemble plans that need legal, financial, and technical input in an order the problem decides, all against a shared workspace.
Investigative Analysis
Work through open-ended investigations where each new finding on the board reshapes which agent should look next.
Expert Panels
Run a panel of specialist agents that contribute to a shared conclusion, with each speaking when the state makes it relevant.
Consensus Building
Drive a group of agents toward an agreed recommendation, halting the loop once the board stabilizes with no further objections.
Dynamic Workflows
Coordinate agents on tasks whose steps cannot be sequenced in advance, letting the shared board decide the flow.
Where Blackboard Multi-Agent Fits
A shared-workspace pattern for agent collaboration
Where debate prompting structures disagreement and branch-solve-merge splits then recombines, the blackboard offers a shared workspace with relevance-based turns. Reach for it when the order of collaboration should follow the problem rather than a script.
Related Techniques
Explore complementary multi-agent techniques
Collaborate on a Shared Board
Explore how a shared workspace can coordinate multiple agents in the Prompt Builder, or see related multi-agent techniques.