Hub-spoke
Dense reasoning core radiating to satellite specialists. Orchestrator + specialists.
Verified against anthropic.com
A hub-and-spoke agent puts a single orchestrator at the centre. It holds the task context, decomposes the request, and dispatches sub-tasks to specialist sub-agents arranged around it, then aggregates their returns into one answer. The specialists never talk to each other; every decision and every write routes back through the hub. In anatomy terms the hub is the reasoning core, and each spoke is a scoped effector or sub-agent bound to a single capability. It is the most common production shape because it keeps control, cost and auditing in one place, at the price of everything depending on the centre.
The hub is the reasoning core; the spokes are specialist sub-agents or effectors bound to one capability each; the policy perimeter wraps the hub's dispatch path, so every write is gated in one place.
- ·Customer-service triage: a router classifies the ticket and dispatches to billing, technical or account specialists (Salesforce Agentforce, Intercom Fin, Decagon).
- ·Multi-tool orchestration: one planner calls search, code and data tools as specialists and fuses the results.
- ·Enterprise workflow routing, where a central agent hands work to department-specific handlers.
- ·Central control and aggregation: one place to reason, gate and audit every action.
- ·Parallel specialist dispatch keeps latency down versus a long sequential chain.
- ·Specialists can be added or swapped without touching the others.
- ·Traceable: every decision passes through the hub.
- ·The orchestrator is a bottleneck and a single point of failure.
- ·Round-trips through the hub add latency and token cost as the fan-out grows.
- ·The hub's context window caps how many specialists it can coordinate at once.
- ·Weak at open-ended autonomy: the hub must anticipate the decomposition up front.
Mapping classified by the Field Book from each agent's documented control flow.
Description, use cases and trade-offs: Field Book analysis, synthesised from established agent-architecture literature.
The corpus does not rewrite vendor documentation. Where copy is quoted verbatim it is marked as such and attributed to its source.