Microsoft Enables AI Agent Features in Windows 11 for Insider Builds

New Windows 11 Dev and Beta Insider builds (26220.7272) have introduced one of the most significant AI-related changes to the OS so far: a new system-level toggle called “experimental agentic features.”

Contents

This option marks a major step toward Microsoft’s vision of a “Windows agentic OS,” where AI agents can operate your PC with awareness, context, and autonomy — performing tasks much like a human would.

What Are “Experimental Agentic Features” in Windows 11?

The new setting is located in Settings > Privacy & Security, serving as a master permission switch for applications that utilise agentic AI capabilities. When enabled, it grants AI agents — Copilot and third-party apps — the ability to interact with your Windows environment using powerful new system-level capabilities.

This includes the ability to:

1. Understand what’s on your screen

Windows exposes a form of system-level computer vision, allowing AI agents to “see” UI elements, text, buttons, images, and application layouts.

2. Click, scroll, and type like a human

The agent can actively control windows and apps:

Clicking buttons
Entering text into fields
Scrolling feeds or documents
Navigating menus
Triggering shortcuts

This behaviour goes beyond traditional macros: agents act dynamically based on content, not fixed instructions.

3. Chain complex actions together

For example, an AI agent could:

Open File Explorer
Locate a folder of photos
Crop them and convert them
Rename them using a pattern
Compress them
Email the package

All without the user needing to manually intervene.

4. Use reasoning to complete tasks end-to-end

Because these agents integrate with large language models, they can interpret high-level natural language instructions and decide which sequence of actions to perform.

Copilot Actions: The First Feature Requiring This Setting

Microsoft has previewed Copilot Actions, an upcoming Windows 11 feature that heavily relies on these agentic capabilities.

Examples of commands you could give Copilot:

“Organise my desktop files by type.”
“Take all these screenshots, crop them, and save them in 16:9.”
“Automate this repetitive task in Excel.”
“Rename all files in this folder using this format.”
“Open this app and finish this workflow for me.”

Copilot will then:

Open the apps involved
Perform the steps visually
Complete the workflow
Report progress
Allow you to watch in real time or minimise it and continue working

But none of this is possible unless you enable the new ‘experimental agentic features’ switch.
Windows otherwise blocks the AI from accessing UI elements and personal data.

Only for Insiders — For Now

This toggle is currently only available to users in the Windows Insider Dev and Beta channels running build 26220.7272. Alongside it, the build adds:

A more natural voice option for Narrator
Updates to the “Click To Do” menu
Several UI refinements

However, the AI agent switch is by far the most transformative feature.

Security & Privacy Concerns

Some testers have already raised concerns after noticing that enabling this feature:

Allows an AI agent to run persistently in the background
Grants access to personal folders
Permits automated input (typing and clicking)
Gives visibility into on-screen content

This level of system access is extremely powerful — but also potentially risky if not properly sandboxed, audited, and permission-controlled.

Microsoft will likely implement granular controls before general release, but for now the toggle acts as a global YES/NO switch.

What This Means for the Future of Windows

This new setting is a clear signal of Microsoft’s long-term strategy:

Windows is evolving from a traditional operating system into an agentic platform — one where AI can perform tasks for users, manage workflows, and bridge cloud intelligence with local applications.

If developed responsibly, this could become one of the most significant UI and productivity shifts since the introduction of graphical user interfaces.