Human-in-the-Loop Comes to Copilot Studio: Exploring the New Request for Information Feature

How do you get human input in the right places in your Copilot Studio agents?

Microsoft has just released a new "Request For Information" (RFI) feature you can add to Agent Flows that allows you to pause execution of your agent and wait for human input to move forward.

This is a valuable update that goes along with other Power Automate actions like "Post an Adaptive Card to Teams and wait for a response" that allow us to collect input as part of flows and act on those responses.

What is still missing though is the seamlessly integrated human input approach that was shown as part of announcing Autonomous Agents nearly a year ago (see the video section in the comments).

Working out where your agents should be supervised and helped by humans is an essential aspect of building responsible AI tools. If you've built an autonomous agent that just does a task end-to-end without the possibility of throwing out a request to a human, then you've probably introduced unnecessary risk into your process.

New tools like RFI will help in this journey of allowing agents and humans to work together, but there are still big opportunities make this interaction more seamless and to make our Copilot "UI for AI" less clunky in this regard.

How would you like agents to interact with humans? Are your human-in-loop ideals just really hard to build or completely impossible right now?

Posted on Linkedin on 08/24/2025 -> View Linkedin Post here