AI coding glossary
Agent Orchestration
Also known as: agent orchestrator, multi-agent orchestration
In one sentence
The layer that plans, dispatches, and supervises one or more AI agents, deciding which subagent runs when, how outputs flow between them, and when to halt.
Full definition
Agent orchestration is the layer above individual agents that decides which subagent to spawn, in what order, with what inputs, and when to halt. It's the difference between 'one chat with an LLM' and 'a team of LLMs working on a feature.' In 2026 the canonical orchestration patterns are: fan-out/fan-in (parallel parallel work, merge results), pipeline (sequential stages with isolated context per stage), specialist-plus-critic (one agent produces, one critiques in a fixed loop), and sandbox-then-merge (subagent works in a git worktree, parent reviews diff). Claude Code's sub-agents and Antigravity 2.0's multi-agent architecture both ship with orchestrators built in; LangGraph and CrewAI are the popular open-source orchestration libraries.