You need a multi-agent pattern where a central orchestrator interprets user goals, delegates to specialist agents (billing, network, devices), and synthesizes a single response for an ISP support desk. Specialists should not freely chat with each other. Which orchestration pattern should you implement?
Select an answer to reveal the explanation.
Short Explanation
You want hub-and-spoke, also called orchestrator-subagent, because the hub can interpret the user's goal, delegate to billing, network, and device specialists, then merge their answers into one response. It keeps specialists from chatting freely and gives you one clear control point for planning, guardrails, and synthesis.
Full Explanation
AI-500 multi-agent patterns include hub-and-spoke, sequential, parallel, peer-to-peer, and orchestrator-subagent. ISP support with distinct skill domains fits hub-and-spoke: the hub owns dialogue state and policy, spokes run bounded tasks, and the hub merges outputs. This simplifies guardrails and logging versus fully connected peers. The unstructured peer-to-peer mesh where every agent messages every other agent without a coordinator is wrong: it complicates control and audit and violates the ban on free specialist chatter. Peer-to-peer is valid for other collaborations, but not here. The strict single-agent design that refuses specialists underuses specialization. Randomly spawning agents without an orchestration graph or aggregation step is not a reliable pattern and lacks controlled planning or synthesis. Sequential or parallel hybrids can sit under the hub for multi-step tickets. Use Agent Framework or LangGraph to encode the graph, timeouts, and aggregation logic explicitly.