Skip to main content
A workflow puts an agent to work through a Run Agent step: the step runs your agent over the workflow’s data and produces a result the later steps can use. It’s how a workflow reasons — summarizing a call, drafting a recap, assessing a deal — instead of just moving data around.

​
The essentials

  • Publish the agent first. A Run Agent step points at an agent by id, and a workflow can’t publish while it references an agent that’s still a draft. Build and publish the agent, then reference it. See Build an agent.
  • Add the step and write its instruction. In the workflow, add a Run Agent step, choose your published agent, and tell it what to do this run. The instruction can pull in data from earlier steps.
  • Use the result downstream. The agent’s output is available to later steps, so you can post it to Slack, drop it into an email, or write it to your CRM.
For the full step reference — the instruction field, how the output is exposed to later steps, and the publish-the-agent-first rule — see Run an agent from a workflow.

​
In practice

A meeting ended workflow runs your published call-summarizer agent to summarize the call and list action items, and the next step posts that summary to the account’s Slack channel — the reasoning and the delivery both happen automatically in one run. Questions? We’re here to help at support@clearskies.cc