noun · AI architecture
Orchestration
also called workflow
The conductor. It decides which AI call or tool goes next, what each one is handed, and what happens when one fails.
The code or controlling agent that coordinates multiple model calls, tools and agents, including their order, inputs, retries and handoffs.
Ordinary program logic, or a controlling agent, that sequences model calls, tools and sub-tasks, and decides what text passes between them. Because each model call is stateless, anything one step needs from another must be passed explicitly.
Real-life example
You build a tool that writes a report. One step searches, one step drafts, one step checks facts and one step formats. A small program calls them in order and passes each result to the next. When the fact-check step fails, the program retries it instead of publishing.
Where the analogy breaks
-
Orchestration is often plain code, not another AI. A fixed sequence with checks can be more reliable than letting a model decide every step.
-
More agents is not automatically better. Each handoff is text, so detail gets lost and mistakes can pass down the chain.
-
Failure handling is half the job. Steps time out, return bad output or loop, and the orchestrator must decide to retry, stop or ask a person.