The AI Agent workflow action helps you automate multi-step tasks inside Workflows using plain-language instructions and selected tools. Instead of building every step manually, you configure one AI Agent action to decide what to do, which tools to use, and what inputs to provide during execution. This article explains how the action works, how to configure it, and how to review its output.


Note: AI Agent is a premium workflow action and may require add-on access to use.

TABLE OF CONTENTS

What is the AI Agent Action in Workflows?


AI Agent is a Workflow AI action that plans and executes tasks autonomously based on your instructions and the tools you make available to it. It’s useful when a workflow needs flexible decision-making, dynamic tool usage, or multiple actions completed in sequence without manually building each step.


For example, instead of creating separate workflow actions to find a contact, generate a booking link, format a date, and send follow-up messages, you can configure one AI Agent action to handle the full process.

Key Benefits of the AI Agent Workflow Action


  • Autonomous execution: The agent decides which tools to use, in what order, and with what inputs based on your instructions.

  • Faster workflow building: One AI Agent action can replace multiple manually configured workflow steps for supported use cases.

  • Flexible configuration: You can build from scratch or start with a template and customize the setup.

  • Downstream workflow support: The agent can return text or JSON output for use in later workflow steps.

  • Built-in visibility: Execution Logs let you review the agent’s reasoning steps, tool usage, and token consumption. 

AI Agent Action Details


These settings control how the AI Agent behaves and provide the core inputs needed to configure the action correctly:


  • Action Name: Add a custom name for the action. The default name is AI Agent, but renaming it based on the use case makes the workflow easier to manage.

  • Template: Choose a pre-built template to prefill the Instructions and Tools fields, or select Build Your Own to start from scratch. Some templates require a specific workflow trigger.

  • Instructions: Define what the agent should do, how it should make decisions, and any rules it should follow. This is the most important part of the configuration.

  • Enhance Prompt: Improves loosely written instructions by restructuring them into a more detailed and organized prompt. You can undo the enhancement if needed.

  • Model: Select the AI model that powers the agent.

  • Tools: Choose the actions the agent is allowed to execute.

  • Conversation Memory: Uses a rolling summary of past runs for the same contact on this step so the agent has context.

  • Output Format: Choose how the agent formats its response.


Templates


Templates help you get started faster by pre-filling common instructions and tools for specific workflow scenarios. They’re useful when you want a starting point that matches a known use case and then customize it for your business.


Templates are pre-built agent configurations that populate the Instructions and Tools fields. If a template requires a specific trigger and that trigger is not already present, the template appears unavailable until the required trigger is added.


Available templates include:


  • Build Your Own
  • Form Lead Follow Up
  • No Show Appointment Recovery
  • Facebook Lead Nurturing
  • Stale Deal Nudge Agent
  • Lead Research & Enrichment
  • New Appointment: Enrich & Confirm
  • Quiz Lead Scoring & Routing
  • Call Transcript Summary & Action Items
  • Task Creation from Won Deals
  • Instagram Comment-to-DM
  • Facebook Comment-to-DM
  • Lead Pipeline Tracker

Writing Instructions


Strong instructions help the agent make better decisions, use tools correctly, and produce more reliable results. The clearer your guidance is, the easier it is for the agent to carry out the task as intended.


The Instructions field defines what the agent does, how it reasons, and how it uses the tools you provide. You can keep instructions simple or make them more detailed depending on the complexity of the task.


Good instructions usually include:


  • A role statement
  • Context about the available trigger data
  • A step-by-step workflow
  • Decision logic using if/then rules
  • Messaging guidelines
  • Rules or constraints for what the agent should avoid doing


If you’re unsure how to structure the prompt, enter your instructions in plain language and use Enhance Prompt to reorganize them.


The editor also supports rich text formatting, including headings, bold text, bullet lists, numbered lists, code blocks, and block quotes.


Configuring Tools


Tools determine what the agent is allowed to do during execution. Giving the agent the right tools, and deciding which fields it can control, helps balance flexibility with predictability.


Click + Add Tool to open the tool picker. Tools are organized by category, and you can also search by name. Each selected tool appears in the configuration panel and can be configured individually.


When configuring a tool, you can choose how field values are populated:


  • Let AI decide all field values: ON: The agent generates values for all fields at runtime based on your instructions and workflow data.

  • Let AI decide all field values: OFF: You can manually set some values yourself while allowing the agent to fill any remaining blank fields.


Each field also includes its own toggle, so you can switch individual fields between AI-generated and manually configured values.


How to Set Up the AI Agent Workflow Action


A structured setup makes it easier to choose the right template, write clear instructions, and give the agent the tools it needs to complete the task successfully.


  1. Go to Automations > Workflows.



  2. Create a new workflow or edit an existing one.



  3. Add a relevant trigger like Form Submitted, Contact Created, etc.



  4. Click + Add (or select the Please select action placeholder) in the workflow canvas.

  5. Select AI Agent from the Workflow AI section.

  6. Enter an Action Name.

  7. You can select Build Your Own to configure the action from scratch.

  8. Add your Instructions.

  9. Click Enhance Prompt to improve the structure of the prompt.

  10. Choose the Model you want to use.

  11. Click + Add Tool and select the tools the agent can use.

  12. Configure each tool, including whether AI should decide field values or whether specific values should be set manually.



  13. Or you can also select a Template: Templates are pre-built agent configurations that populate the Instructions and Tools fields. If a template requires a specific trigger and that trigger is not already present, the template appears unavailable until the required trigger is added.



  14. Open Advanced Options (optional) to enable Conversation Memory or configure Output Format:

  • Conversation Memory: Helps the agent keep continuity across repeated runs for the same contact on that step. This is helpful for recurring workflows where the same contact may re-enter the step and the agent should adapt based on previous interactions.

    • Off by default: When enabled, the agent uses a rolling summary of past runs for the same contact on that step.

    • Use this for recurring or multi-touch workflows (for example, periodic follow-up). For one-time workflows (for example, a single form follow-up), it can usually remain off.

  • Output Format: Controls whether the agent returns data that later workflow steps can use. Choosing the right output type helps when the workflow needs to reference the agent’s result after it finishes running.

    • None: Performs actions but does not return structured output for downstream steps.

    • Text: Returns a free-text response. When selected, an Output Description field appears so you can describe what the response should contain.

    • JSON: Returns structured data based on a defined schema. You can add properties with a name, type, and optional description to make the output easier to reference later in the workflow.

  1. Save the action.

Execution Logs


Execution Logs make it easier to understand what the agent did during a workflow run. They are useful for validating behavior, checking tool usage, and troubleshooting unexpected results.


After the workflow runs, open the Execution Logs tab in the workflow builder and select an AI Agent execution to view the trace.


The trace can include:


  1. Agent Input: The full input received by the agent.

  2. LLM: Each model call used to reason about the next step.

  3. Tool executions: Each tool the agent called, including execution time and success or failure status.


You can expand each step to review input and output data. The log view supports both a table format and raw JSON view.


Frequently Asked Questions


Q: Is there a limit on how many tools I can add?
Yes. You can add up to 10 tools per agent. Internal tools such as Date Calculator and Math Operations do not count toward this limit.


Q: Which model should I use?
Use a general-purpose model for most use cases, choose higher “thinking” levels for more complex logic, and choose lightweight models for simpler tasks where speed matters.


Q: Can I use the AI Agent with any trigger?
Yes. AI Agent works with any workflow trigger. Templates may recommend specific triggers because their instructions are written for that context.


Q: What does Enhance Prompt do?
Enhance Prompt restructures your instructions into a more organized and detailed prompt, including clearer steps and edge-case handling. You can undo the enhancement if needed.


Q: When should I use the JSON output format?
Use JSON when later workflow steps need to reference specific values from the agent’s response, such as qualification results or other structured outputs.


Q: How can I see what the agent did during a workflow run?
Open the Execution Logs tab in the workflow builder and select the AI Agent execution to review the trace, including model calls, tool executions, inputs, outputs, and token usage.


Q: Do templates lock me into a specific configuration?
No. Templates pre-fill the Instructions and Tools fields, but you can edit the instructions, change the model, add or remove tools, and update advanced options after applying a template.


Q: When should I enable Conversation Memory?
Enable it when the same contact may go through the agent step more than once and the agent should adapt based on prior runs. For one-time workflows, it is usually not needed.