Copilot Is About to Click Buttons for You. Who Decided What It's Allowed to Touch?
October 13, 2026
On October 7, Microsoft said Copilot on Windows will soon do more than answer questions. On Copilot+ PCs, Copilot will be able to understand content on your PC, including files and recent activity, if you give it permission. It will also be able to take action for you, such as organizing files, checking device diagnostics, troubleshooting issues and carrying out workflows right on the device. Microsoft expects these features to start rolling out "in the coming months."
That's a real change. For the past couple of years, most of us have used AI like a smart search box: you ask, it answers, and you decide what to do next. Now the assistant is starting to do the next step itself.
From chatting to doing
This didn't come out of nowhere. On September 25, Microsoft introduced a new Copilot built around three experiences: Home, Code and Autopilot. Inside Home, a mode called Cowork handles "work you delegate": you define a task and it runs end to end. Microsoft also announced Office in Copilot, which can create or update real Word, Excel and PowerPoint files. And Autopilot is described as "a persistent, proactive and personal agent that keeps working even when you're not." It can follow up on threads, run recurring work and reach out to people for updates.
Put those together and the picture is clear. AI is moving from something you talk to toward something that works alongside you, opening files, editing documents, sending follow-ups and changing settings.
The part nobody asked about
Microsoft is clearly thinking about control. The October 7 announcement leans hard on security. It covers what an agent can access, knowing which agent did what, and managing it all through tools like Intune. It also introduced Microsoft Execution Containers, which let organizations decide which files and networks agents can reach.
That's good news. But every one of those controls is a setting, and somebody has to set it.
In most small businesses we work with, that's where things get uncomfortable. Not because anyone did anything wrong, but because businesses grow organically:
- A shared drive where "everyone has access" because that was easier five years ago.
- A former employee's account that still owns a few important folders.
- A process that only works because Linda knows to click the third button, not the second.
- Admin rights handed out to fix one printer problem and never taken back.
When a person works inside that kind of setup, their judgment fills in the gaps. They know which folder not to touch and which customer file is sensitive. An AI agent acting for you works with whatever access you have. If your permissions are loose, its reach is loose too, and it moves a lot faster than Linda does.
Helpful AI starts with a map
None of this is a reason to avoid these tools. Done right, an assistant that can organize files, draft the document or chase down the follow-up is a real gift to a small team that's stretched thin.
It only helps, though, when someone has done the unglamorous work first:
- Knowing what it can see. Which files, mailboxes and systems would Copilot actually reach for each person? That's usually a bigger list than owners expect.
- Deciding what it can do. Some actions are low risk, like tidying a downloads folder. Others, like emailing a client, changing a financial spreadsheet or touching device settings, deserve a human approval step.
- Matching it to how you actually work. Every business has its sticky workflows, the ones held together by habit and a few people's memory. Agents should fit around those on purpose, not stumble into them.
- Keeping an eye on it. When something goes sideways, you want to know whether a person or an agent did it. Microsoft built agent identity in for exactly that reason, but it only pays off if someone is watching.
That's the heart of our AI advisory work. We sit down with you, map how work and data really flow through your business, and put sensible guardrails in place (permissions, data access and approvals) before an agent starts clicking buttons. As a Microsoft Certified Partner, we follow these rollouts closely so you don't have to track every announcement yourself.
Last week we wrote about employees already using AI on their own. This is the other side of that coin. Soon AI won't just be something your team uses. It will be something acting for them, and the time to decide what it's allowed to touch is before the rollout reaches your desks.