Coordinating multiple AI agents and tools to work together to complete multi-step tasks.
Also known as: Agent Coordination, Multi-Agent Orchestration, Orchestration of Agents
In plain English
Agent orchestration is like a project manager for AI: it coordinates different AI 'helpers' and apps so they work together. For example, one helper finds data, another writes a draft, and orchestration passes results between them and handles errors.
In practice
In practice, agent orchestration appears as platforms or workflows that chain models, retrieval systems and external tools to automate business tasks. It matters because it affects reliability, latency, cost and compliance — you evaluate connectors, monitoring, retry policies and guardrails when selecting or building a solution.
Under the hood
An orchestrator implements a control plane that decomposes goals into subtasks, schedules agents, routes messages/state, and handles retries, timeouts and failures. Key concepts include planners (task decomposition), executors/tool adapters, message-passing protocols, shared context (memory or vector stores), observability and safety modules; integrations often rely on function calling and retrieval-augmented generation.
Example
"We should use agent orchestration to coordinate the data lookup, summarisation and email-sending agents for the weekly report."