How does the AI agent step work in a workflow?

The configurable agent step runs an AI agent inside a workflow—you write plain-language instructions, assign skills, and the agent completes the task, pausing where it needs human input.

The configurable agent is the most powerful step type in a workflow. Instead of a fixed list of actions, you describe what you want done in plain language and the agent figures out how to do it—using the skills and tools you assign.

How to configure it

When you add a configurable agent step, you write:

Goal: what the step should achieve overall (e.g. "Find the employee's probation end date and draft a notification to their manager").
How: constraints on the approach (e.g. "Use the HRIS skill to look up the date. Draft the email but do not send without approval").
When to stop: the condition that tells the agent its job is done (e.g. "Stop when the draft has been approved and sent").
On error: what to do if something goes wrong (e.g. "Post an internal note on the ticket and stop").

You also assign skills—the toolsets the agent can use to complete the task.

What the agent can do

Depending on which skills you enable, the agent can:

Search your knowledge base and company documents.
Look up employee data from a connected HRIS.
Create or update tickets, add internal notes, send ticket replies.
Send email or Slack messages (with approval, if configured).
Call external tools registered by IT via MCP integrations.
Pause and wait for a human reply before continuing.
Delegate a sub-task to another agent (for example, to gather information from a third party).

Skills and tool loading

The agent does not call every assigned skill at once. It loads skills on demand as it decides they are relevant, and remembers what it has already loaded across pauses and resumes. This keeps its behavior focused and efficient.

Example

Goal: "Prepare the leaver's exit pack."
Skills enabled: HRIS lookup, Document generation, Ticket management.

Steps the agent takes automatically:

Looks up the employee in HRIS to get their last day and manager name.
Generates the exit documentation with those details.
Creates an internal task on the ticket for payroll to review.
Posts a summary reply on the ticket confirming it is done.

Guardrails

The agent follows your instructions—write them precisely. Vague goals produce unpredictable paths.
Enable only the skills the step needs; unused skills are extra surface area with no benefit.
Use approval steps (see the approval article) when the agent's actions have irreversible consequences such as sending external email or writing to another system.
Test in a pilot inbox before enabling on live traffic.