Multi-agent workflow
A multi-agent workflow is a job split across several AI agents, each doing one part and handing work to the next. One agent may read a contract, another may review its clauses, and a third may write the summary.
In a run
In the document review scenario, agents take in a contract, review its clauses and write a summary for the client. Each handoff goes through one controlled path, and the summary is checked before it leaves the network.
Our word for it
The workflow.
Why handoffs matter
Each handoff is a place where access can grow by accident. If the second agent holds more than the first, a request can reach further than the person who started it intended.
On Ares the new permission is the overlap of what the caller holds and what the next agent needs. It gets smaller or stays the same. It never grows.
The rule holds across companies too. When work crosses to a supplier, the permission narrows at the company boundary, and both records show what crossed. See across organisations. Every handoff is on the record, so any step traces back to the person or event that started the job.
Steps a person takes
Agents are chained into a workflow with defined steps. You choose which steps wait for a person. Creating, changing and running a workflow are separate rights, so the person who runs a workflow need not be able to change it.
Where it stops
The workflow follows the steps and the rules you approve. It does not judge whether a step is wise, and it covers only the agents you enrol.
Run the document review scenario, or browse the workflow library.
Related terms
- OrchestrationAI agent orchestration is the coordination of several agents and tools so they work through one job in the right order.
- DelegationDelegation is one party handing a task, and part of its authority, to another.
Tell us about one workflow.
We will say whether an agent should do it, and what the first step costs.
A senior member of the team replies within two business days.