Microsoft launches Copilot Tasks, an AI with its own computer

AI now has a desktop of its own. Not a metaphorical one, but a real one: a cloud computer with a browser that it uses to go online and do the kind of work that usually eats up hours of routine. That is how Copilot Tasks works, a system Microsoft unveiled as a public preview on Thursday, February 27.

The idea is simple and ambitious at the same time. The user describes a task in plain language, and Copilot Tasks takes it over and runs it in the background without loading the user's own device. Scheduling a meeting, generating a study plan, processing information: all of it happens on Microsoft's cloud infrastructure, on a dedicated virtual machine with its own browser. When the work is done, the system sends a detailed report. Tasks can be run once, scheduled for a specific time, or set to repeat on a regular basis. In effect, this is a full-fledged AI assistant that works on a schedule.

To grasp the scale of what's happening, it helps to recall the context. For the past year and a half, the biggest tech companies have been quietly racing to turn language models from text generators into autonomous agents that can act in the real world. Google is developing Project Mariner, an experimental agent that can interact with web pages. Anthropic released Computer Use for Claude, a feature that lets the model control the cursor and keyboard. OpenAI is working on Operator, an agent that performs tasks in a browser. With Copilot Tasks, Microsoft enters the race with its own trump card: the deepest possible integration into an ecosystem that hundreds of millions of people use every day.

The key difference in Microsoft's approach is architectural: a dedicated cloud computer. Rather than running the agent on the user's device, where it could conflict with other apps and risk leaking data, the company isolates the process in its own cloud environment. That solves several problems at once. First, performance: the AI agent doesn't compete with the user's apps for CPU and RAM. Second, security: the virtual environment limits the potential damage from the agent's mistakes. Third, availability: tasks keep running even when the laptop is closed or the phone is off.

But this is also where the biggest questions lie. How broad is the range of tasks Copilot Tasks can actually handle in practice? Big-company demos traditionally look impressive, but real-world use often exposes limitations. Scheduling a meeting in an Outlook calendar is a clear, controlled scenario. What about more complex chains of actions that require working across several services, making decisions under uncertainty, or handling unusual situations? For now, Microsoft is showing the system as a preview, which means answers to these questions will come with time.

There is also a more fundamental issue: trust. Giving an AI agent the right to act on your behalf, even for simple tasks, demands a fundamentally different level of trust than asking a chatbot to write an email. When an agent works autonomously, in the background, on a cloud computer, the user loses the ability to oversee every step. A report at the end is nice, but what if the agent made a mistake partway through? What if it sent the wrong email or booked a meeting for the wrong time? The industry still has to develop transparency standards and rollback mechanisms for autonomous agents, and those standards are far from mature.

For Microsoft, the launch is a strategic move that strengthens Copilot's position as the centerpiece of the user experience across Windows and Microsoft 365. The company has steadily expanded its AI assistant's capabilities, turning it from an add-on to Bing search into a full platform for automating everyday tasks. If Copilot Tasks lives up to expectations, it could change the basic model of how people interact with computers: instead of switching between apps and doing routine operations by hand, the user simply states an intent and gets a result.

We are on the threshold of an era in which AI stops being a conversation partner and becomes a doer. Copilot Tasks is one of the first mass-market products to embody that shift. The coming months will show how smoothly the move from preview to full release goes and how ready users are to hand real actions over to a machine. But the direction is set, and there is no going back.