Usually, when creating a workflow, every decision has to be predefined as a branch, every condition anticipated, and every field manually mapped. If you’d prefer not to take every single variable into account, you can use the AI Agent action to describe your goal, give the agent access to tools, and it will plan the sequence, decide the inputs, and handle edge cases without the need for manual input. Let’s learn how to use it.

Creating Your Workflow #

• Navigate to the Workflows tab inside the Automation section, and edit or create a workflow.
• Inside the workflow builder, add the action AI Agent.
• Choose a template or select Build Your Own to start from scratch.

Using a Template #

Templates only work with specific triggers, so selecting one will prompt you to apply the trigger and template.

Agent Configuration #

• Write your instructions describing what the agent should do. If using a template, this will be prefilled, but you can edit it as you see fit. Even if you write rough instructions, you can click Enhance Prompt so the AI restructures your input into a production-ready prompt.
• Choose between GPT-5.2, GPT-5.1, and GPT-5 Nano to run the workflow as you prefer. GPT-5.2 Low thinking is the recommended default for most use cases.
• Click Add the tools to set the tools that the agent can use (up to 10 per agent). You can select any workflow action for this, including those related to connected apps.
• Once a tool has been added, you can click the pencil icon to edit it or the X icon to delete it.
• Once satisfied with your tools, open the Advanced Options to enable Conversation Memory, which allows the agent to retain a rolling summary of past executions for the same contact, or Structured Output if you want the agent to return results as Text or JSON with a user-defined schema to aid If/Else branches, field updates, or downstream workflow steps.

Once satisfied, click Save Action and you’re done. The AI Agent transforms workflows into a tool where a true CRM agent can evaluate context, make judgment calls, and execute actions on its own, all within the system you already know.

Frequently Asked Questions #

What does the AI Agent action do? #

It lets you describe a goal and give the agent tools, and it plans the sequence, decides inputs, and handles edge cases without requiring every branch to be predefined.

Do I have to start from scratch? #

No. You can choose a template or select Build Your Own. Templates only work with specific triggers and will prompt you to apply the matching trigger.

Which AI models can I choose from? #

You can choose between GPT-5.2, GPT-5.1, and GPT-5 Nano, with GPT-5.2 Low thinking recommended as the default for most use cases.

How many tools can an agent use? #

You can add up to 10 tools per agent, including any workflow action and actions related to connected apps.

What is Conversation Memory? #

It allows the agent to retain a rolling summary of past executions for the same contact across runs.

What is Structured Output for? #

It returns results as Text or JSON with a user-defined schema to aid If/Else branches, field updates, or downstream workflow steps.