Getting useful results from Copilot agents takes more than describing what you want and letting the agent run. You need to decide what to delegate, define the scope of the task, give the agent the context it needs, and know when to step in and redirect its work.

In this demo-focused session, we’ll use Copilot agents in Visual Studio for code exploration, change planning, multi-file implementation, and testing. Rather than focusing on generating code, we’ll look at the decisions a developer makes along the way: what to delegate, how to set useful boundaries, and how to keep the agent focused on the task.

We’ll also look at how to prepare a repository for Copilot with relevant context and repository instructions. As the agent works, we’ll examine how to recognize when its approach does not fit the task and redirect it. Finally, we’ll review the proposed changes in Visual Studio for scope and correctness using the new review tools, then use the test tools to validate behavior before accepting the work.

You will learn to:

• Decide which development tasks to delegate to a Copilot agent and define clear boundaries for the work.
• Prepare your repository with context and instructions that help guide the agent through the task.
• Redirect an agent when needed, review its proposed changes for scope and correctness, and validate the resulting behavior with tests.