Overview
AI Agents can now include a human approval point in the middle of a flow. Human Validation is a checkpoint you set up inside the agent's own Behavior: at higher-risk moments, the agent pauses and sends its output to a person for approval before continuing. The control lives inside the agent's behavior, with no separate automation to configure.
What's new?
- A checkpoint set up right inside the Behavior: type `/` in the instructions and select Request Human Validation from the Actions menu, at the point in the flow where human review should happen.
- Route the review to a specific person through Send to, with the phase SLA applied automatically.
- The validation appears as its own chip on the behavior, so it's always clear which steps have a human checkpoint and which don't.
How does this help?
- Puts human judgment exactly where the risk is: the agent runs on its own through most of the flow and stops only at the points that require a decision, such as a payment above a threshold, a contract clause, or a customer-facing message.
- Makes it safe to run agents in higher-stakes production processes: you keep end-to-end automation and add an explicit brake where a mistake would be costly.
- Keeps governance visible on the agent itself: the validation chip shows which steps carry human approval, with no need for separate documentation to understand the agent's behavior.
- Uses the phase SLA automatically: the review inherits the step deadline, so the checkpoint fits into the process timing with no extra setup.
Highlight: pause, validate, and continue
The biggest value of Human Validation is keeping the agent running with a real stopping point: the agent runs, requests the validation, and continues once the decision is recorded. You build this pattern by chaining Behaviors.
At the checkpoint, the reviewer fills a field as part of the validation, for example a "Final decision" Radio Select with Approve or Reject options. You set the next Behavior to trigger on A field is updated watching that same field. The next Behavior only runs once the review is complete, and it runs different paths depending on the reviewer's choice. The whole chain runs on its own, but each link only advances after the recorded human decision. The full example is in the Help Center article.
Use cases
1. A purchasing triage agent in an expense approval process
The agent classifies each request by category and amount. You add Request Human Validation for requests above the spending limit and route it through Send to the area manager. Purchases within the limit move on automatically; only high-value ones wait for approval, with no parallel control spreadsheet.
2. A legal response agent in a contract review flow
The agent drafts the response based on the identified clauses. Before the response goes out, the checkpoint sends the draft to the legal team to review sensitive clauses. The contract only advances a phase after sign-off, keeping text with contractual risk from moving on without human reading.
3. A support agent in a customer-facing process
The agent generates the reply to the requester based on the card history. Human Validation routes the message to the operator for approval before sending, ensuring no external communication goes out unchecked, even with the rest of the support flow automated.
Most relevant for
- AI agent builders
- Pipe Admins
- Operations teams
- Purchasing managers
- Legal and finance process teams
Documentation
For the full step-by-step, check the Help Center article.
Plans and availability
This feature is rolling out gradually to all accounts.
