Blackboard
A shared workspace agents read and write opportunistically, without direct coupling. No fixed control flow.
A blackboard agent coordinates through a shared workspace rather than a control flow. Every agent reads the current state of the blackboard, contributes what it can, and writes its result back for others to build on, opportunistically, with no fixed order and no central dispatcher. Coordination is emergent from the shared state: an agent acts when the blackboard holds something it can advance. It suits open, ill-structured problems where the path is not known up front and different specialists each move the work forward when they are able. In anatomy terms the memory plane is the coordination substrate, and the reasoning cores orbit it.
A shared memory/state plane (the blackboard) is the coordination substrate; multiple reasoning cores read and write it opportunistically; there is no orchestrator; the perimeter governs what may be written back.
- ·Open, ill-structured problems solved incrementally by several specialists.
- ·Sensor-fusion or multi-source analysis where contributions arrive out of order.
- ·Design or planning tasks where the path emerges from the shared state.
- ·Handles ill-structured problems with no predetermined path.
- ·Specialists stay decoupled: they coordinate through state, not messages.
- ·Add a new contributor without touching the others.
- ·Resilient: no central dispatcher to fail.
- ·Emergent control is hard to predict, bound and audit.
- ·Contention and consistency on the shared state are real problems.
- ·Can stall if no agent can advance the current state.
- ·Termination is harder to guarantee.
- Internal research (Scalarity), Agents vs Operators / Agent Archetypes v2 · unvalidated
Description, use cases and trade-offs: Field Book analysis, synthesised from established agent-architecture literature.