Multi-agent workflows
On this pageWhat it is
What it is
A multi-agent workflow uses more than one AI agent on a single job, each with its own role and instructions. One agent might research a lead, a second might write an email and a third might check the draft against your tone of voice before anything is sent. They hand work to each other, or a coordinating agent decides who does what. Botpress, ChatGPT, Claude, Dust, Lindy, Microsoft Copilot Studio, Relevance AI and Zapier Agents are listed with this feature.
Why it matters
Splitting a job into steps often gives more reliable results than asking one assistant to do everything at once, and it lets you test and fix each step on its own. You need it for processes with several distinct stages, such as lead handling, reporting or content production. For a simple task like answering one type of question, a single agent is easier to run and cheaper to keep an eye on.
What to check
- Ask how the agents pass information to each other and whether you can see each step.
- Test a workflow with a deliberate mistake in the first step to see whether later agents catch it.
- Check whether a person can approve or stop the work at a chosen point.
- Find out how usage is counted, because several agents on one task can use far more than one.