What is AI orchestration?
AI orchestration is the control layer that sequences steps, routes work between models, tools and people, handles retries and errors, and keeps state across a multi-step workflow.
In more detail
Models are one ingredient. The orchestrator decides which step runs next, what happens on a timeout, how long to wait for an approval and how to resume after a failure. It can be a workflow engine such as n8n or Temporal, an agent framework such as LangGraph, or plain code. The key choice is how much of the sequence is fixed and how much the model decides. Fixed paths are easier to test and audit. Model-chosen paths handle messy input. We usually fix the outer flow and let the model decide inside a step. Good orchestration also records every run, with inputs, decisions, tool calls and outputs, so you can replay a failure instead of guessing.
Example
A month-end close run waits for bank files, triggers reconciliations, pauses for the controller's approval on adjustments over a set amount and retries a failed ERP call twice before alerting a person.