Skip to main content

AI Agents: Requesting Human Validation

Learn how to add a human review step to an AI agent so a person checks the agent's output before it moves forward.

Written by Product Team

👤 For super admins and admins

🔐 Available on all plans

🎯 For those who want to keep a person in the loop on what an AI agent produces

What this solves:

  • Puts a human check on AI output

  • Routes the review to a specific person, with the phase SLA applied automatically

Looking for human validation in Automations, not AI Agents? See Implementing human-in-the-loop in critical automations.


Before you start:

⚠️ Important: These triggers are not available for human validation:

  • A recurring activity

  • An alert is triggered

  • A card leaves a phase

If one of these is selected, the human validation action appears disabled. If you add human validation before choosing a trigger, these triggers appear disabled in the trigger list.


Step-by-step: How to set up Human Validation

1. Go to your AI agent and open the behavior you want to add the review to.

2. In the behavior instructions, type / to open the dropdown.

3. Under Actions, choose “Request human validation”.

4. In the modal, fill the Title field. This names the validation the reviewer receives.

  • 4.1. Click the button "+” and pick a card field or attribute to build the title automatically.

  • 4.2. Or type the title manually.

5. In Send to, select or type the email of the person who should review the output.

⚠️ Important: If the reviewer is outside the pipe, your company, or Pipefy, activate form sharing for the phase tied to this validation. Without it, they can't open the task.

6. Click Save. The human validation chip is added to the behavior's instructions.


How the reviewer receives the validation

Every time this behavior runs, the person you assigned gets the review as a task.

1. On the Pipefy home page, go to the Tasks and Requests area.

2. Opening it lists the pending tasks, for example: Expense report validation.

3. The reviewer also gets an email, with a button that links straight to it.

When the reviewer clicks on the task, they see the form, enter their decision, and click the submit button.

The task opens the form of the phase tied to your trigger, and follows the phase SLA.

ℹ️ Tip: The reviewer decides based on what's in the phase form, so give them enough context there. Add a dynamic content field to the top of that phase form to show instructions, links, or data pulled from the card.


Which form the reviewer sees

The task always opens a phase form. Which phase it uses depends on the behavior's trigger:

Trigger

Phase form the reviewer sees

A card is created

The pipe's first phase

A card enters a phase

The phase the card enters

A field is updated

The phase where that field lives

An email is received

The phase where the email arrives

All connected cards are moved to a phase

The destination phase


How to chain behaviors after a validation

If your agent has more than one behavior, you can make the next behavior wait for the human validation to be completed.

💡 The idea is: use the field the reviewer fills to trigger the following behavior.

Example:

Behavior 1 requests human validation. The validation phase has a radio button field named Final decision, with the options Approve and Reject.

To make Behavior 2 run only after the review:

  1. In Behavior 2, set the trigger to A field is updated.

  2. Choose Final decision as the field to watch.

  3. Continue the behavior setup as usual.

Behavior 2 now runs only when the reviewer fills Final decision, which happens once the validation is done.

Did this answer your question?