What is agent orchestration?
Agent orchestration is the coordination of the steps, tools and hand-offs that one or more AI agents use to complete a task, including the order things happen in and what to do when a step fails.
Why it matters
Most real requests take several steps. Verifying a customer, checking a policy, making a change and sending confirmation must happen in the right order, and a failure halfway through should not leave things half done. Orchestration provides that structure. It also keeps a record of each step, which makes an agent’s behaviour easier to review and correct.
How it works
An orchestration layer sits around the model. It holds the state of the task, decides which agent or tool acts next, passes results from one step to the following one and applies rules such as retry limits and approval checks. Some steps are fixed in advance and others are left to the model’s judgement. If a step cannot be completed, the layer routes the task to a person.
In practice
A customer asks to move a delivery to Friday. The orchestration layer first runs an identity check, then calls the courier’s system for available slots, then asks the customer to pick one, then books it and sends a confirmation message. When the courier’s system does not respond after two attempts, the layer stops, opens a ticket with the details gathered so far and tells the customer.
