Orchestration
The layer that decides which model, tool, or agent runs when, and how their outputs flow together.
Think of it like
A conductor — not playing any instrument, just cueing each section in at the right moment.
Example
A support system routes a refund question to a billing agent, waits for its answer, then hands it to a drafting agent to write the reply.
How it actually works
Orchestration is the plumbing above the models: sequencing steps, passing state between them, retrying failures, enforcing limits. It can be rigid (a fixed workflow you coded) or dynamic (a model deciding the next move each turn). Most production “AI apps” are 20% model and 80% orchestration — the boring glue is where reliability actually comes from.
For product teams
Where a demo becomes a product — the coordination that makes many parts behave like one.
For engineers
The control layer that sequences model calls, tool executions, and agent handoffs, managing state and error handling between them.
Related
- Multi-Agent — A common orchestration shape: many specialized agents.
- Handoff — Passing control from one agent to another.
- Agent Loop — The repeating cycle it often coordinates.
Read anything AI without the jargon
Look up any term in plain English, or save terms as you read with the free Chrome extension.
Open DecoderAdd to Chrome