Overview of the Steps You Can Add to an Agent Team in GTM One

Learn which step types you can add to an agent team and how each one runs.

Overview

An agent team is a sequence of steps that runs from top to bottom. Each step does one kind of work. Some steps run an AI agent, some take an action such as sending a message, and some control the flow by repeating steps or choosing between paths. Knowing what each step type does helps you design an agent team that does the right work in the right order.

Build an agent team with steps

1. In the left navigation panel, click Agent Teams, open the agent team you want to edit, and click the Steps tab.
2. At the bottom of the steps list, click Add action. In the Add action panel, type in the Search actions box or click a tab (All, General Purpose, Orchestrate, Connect, Notify, Output or Retrieve), then click the step type you need.
3. Select the step to open its settings, then fill them in.
4. To repeat steps or split the flow, add a For Each step or a Condition step, then click Add subaction to add the steps that belong inside it.
5. While you have changes that are not saved, the editor shows You have unsaved changes. Click Save changes to keep them, or click Discard to restore the last saved version of the agent team.

The trigger

Every agent team starts with a trigger. The Run as Form trigger collects the inputs a person enters when they run the agent team by hand, and it cannot be deleted. Click Add trigger to add triggers that start the agent team automatically, such as a schedule, a CRM event or a webhook.

Step types

The Add action panel groups step types by task. Some step types are available only on some accounts.

General Purpose

• AI Agent: Uses a flexible AI agent that can take many actions based on your instructions.

Orchestrate

• For Each: Does something for each item in a list.
• Condition: Runs different steps depending on a condition.
• Route to User: Assigns every row in a table to a user by the method you choose.
• Wait: Pauses the workflow for a period of time.
• Human Approval: Pauses the run until a person you choose approves it before the next step runs.
• Run agent team: Runs another agent team using inputs from this one.

Connect

• Connector: Connects to external systems with custom integrations.
• Export: Sends data to an export destination.

Notify

• Email: Sends an email to one or more recipients.
• Slack: Posts a message to a Slack channel or person.
• Activity Feed: Posts an update to the ZoomInfo activity feed.

Output

• Create HTML: Generates a formatted HTML page from the run.
• Create DocX: Generates a Word document.
• Create Excel: Generates an Excel workbook.
• Create CSV: Generates a CSV file from a table.
• Create Audience: Creates a new Audience from the rows in a table.
• Update Audience: Adds or removes records in an existing Audience.
• Update CRM Record: Writes values back to records in your CRM.

Retrieve

• ZoomInfo Agent: Runs a ZoomInfo agent, such as Meeting Prep, and uses its answer in later steps.
• ZoomInfo Record: Looks up a single person or company record in ZoomInfo.
• ZoomInfo List: Finds a list of companies or contacts that match your criteria.

AI Agent steps

An AI Agent step runs an AI agent with the instructions you give it.

• You can base the agent on one of the available agent templates or create a custom agent with its own name.
• By default, an agent uses the global LLM configuration. You can switch it to use its own model settings instead.
• If you change the agent for a step that already has instructions, the editor asks you to confirm, because the change may replace the instructions.

Before an agent team runs, the editor checks the connection status of the tools and MCP servers your steps use, so you can see when a connection needs attention.

What a step takes in and gives back

Most steps have an Input source: the data the step works from. You can pick one or more of:

• User Inputs, the values the run started with (a form, an audience row or a CSV row).
• Agent Outputs, the output of an earlier step.
• The current item of a For Each step, when the step sits directly inside that loop.

To point an instruction at an earlier step's output, write the step's number (e.g., Step 2) in the instruction. The builder shows it as a step reference and renumbers it when you move steps. A Condition step reads only what its input source already holds and retrieves no new data.

For Each (loop) steps

A For Each step repeats the steps inside it once for each item in a list, such as each record in a table.

• In a run, each pass through the loop is listed by number, followed by the item name when one is available (e.g., #1 followed by the item's name).
• In the runs table, the output column for a step inside a loop shows the output from the first item in the loop.
• The run's output view shows up to 100 loop items. When a loop has more, the rest are left out of the view with a note that other loop items have been omitted.

Condition steps

A Condition step sends the flow down one of several paths based on conditions you define.

• Each path has a name and a condition. In the editor, a path without a name is shown as Condition 1, Condition 2 and so on.
• You can add up to five conditions, plus the default path. Conditions should be mutually exclusive, because a record that matches more than one follows the first match.
• The last path is the default path. It applies when no conditions match.
• In a run, the step shows which condition matched (e.g., Condition 1 followed by the condition), or Default path taken - no conditions met.
• Steps on a path that was not taken are marked as not run.

You can place a For Each step inside a Condition path, or a Condition step inside a For Each step, to build more detailed flows.

How steps are numbered in run results

The runs table labels each output column by the step's position:

• A top-level step reads Output: Step 1.
• A step inside a For Each step reads Output: Step 2.1, meaning the first step inside step 2.
• A step inside a Condition path reads Output: Step 2/Condition 1/Step 1.
• The Condition step itself reads Output: Step 2 Conditional routing.

When you add, remove or reorder top-level steps, references such as "step 2" in an agent's instructions are updated automatically to match the new numbering.