Microsoft is taking Copilot from “that chatbox you open sometimes” to something much more ambitious: an AI that quietly works through your to‑do list on its own computer in the background.

The new feature is called Copilot Tasks, and Microsoft is framing it as the second chapter of AI – a shift from chat to actions, from answers and drafts to actual completed tasks. Instead of you hopping between tabs, apps, calendars, and inboxes, you tell Copilot in plain language what you need done, and it goes off to do the grunt work on a separate cloud‑hosted machine with its own browser.

Right now, Copilot Tasks is in a limited research preview, rolling out to a small group of early testers via a waitlist, which is very on brand for how big AI features are being shipped in 2026. Microsoft explicitly says this first phase is about watching how people actually use it, tightening the safety screws, and refining the product before a broader launch.

So what does this thing actually do? Microsoft’s own examples are deliberately everyday and a bit boring – which is exactly the point. Think of the kind of nagging admin you put off: every evening, Copilot can surface only the urgent emails in your inbox, draft replies you can quickly approve, and automatically unsubscribe you from promos you never open. Every Friday, you can watch new apartment rental listings in your area and go as far as booking showings for the promising ones. On Monday mornings, it can drop a briefing into your lap that pulls together your upcoming meetings, travel, and how your calendar actually lines up with the priorities you said you care about.

It isn’t just time management, though. Copilot Tasks leans hard into document generation and research‑heavy chores. A professor or student can hand in a syllabus and get back a full study plan, complete with practice tests and focus blocks scheduled before each exam. A job seeker can continuously track new listings that match their experience, then auto‑tailor their resume and cover letter for each role it finds. For office workers, it can mine emails, attachments, and even images in your inbox, then spit out a polished slide deck with charts and talking points – the kind of task that normally eats an entire afternoon.

The lifestyle examples are where it really starts to feel like an AI “agent” rather than a chat assistant. You can ask it to plan a birthday party end‑to‑end: find and book a venue, send invites, and track RSVPs. You can have it keep an eye on used car listings 24/7, contact dealerships, and line up a test drive when something promising appears. It can hunt down top‑rated plumbers near you, compare quotes, and book the one that makes sense. For travel, it can reserve a ride that’s timed to your flight and automatically adjust if your flight is delayed, monitor hotel rates and rebook if the price drops, and even audit your subscriptions, highlighting the ones you don’t use and cancelling them.

Under the hood, Microsoft is doing something quite different from just “making the chatbot smarter.” Copilot Tasks runs on its own cloud‑based computer and browser, separate from your laptop or phone, so when you say “watch these sites and send me a weekly report” or “compile competitor pricing every Friday,” it doesn’t need you to keep anything open. You give it a goal in natural language, it breaks that into steps, navigates across apps and websites, fills forms, sends emails, and then reports back with what it actually did. Tasks can be one‑off, tied to a specific schedule, or set to recur automatically, depending on how you define them.

This is very much Microsoft planting a flag in the “agentic” AI race. Over the past year, we’ve seen a wave of AI systems that don’t just answer questions but actually operate software on your behalf – from browser auto‑browsing modes to tools that control full desktop sessions in the cloud. Copilot Tasks is Microsoft’s take: an agent that lives in the Copilot ecosystem and uses a cloud PC instead of hijacking your local machine. It builds on earlier efforts like Copilot Mode and Actions in Edge, which could already do things like navigate pages, open tabs, and help book reservations inside the browser.​

But once you let an AI loose with a browser and your data, the obvious question is: how do you keep it from going rogue? Microsoft is trying hard to position Copilot Tasks as a copilot, not an autopilot. The system is designed to ask for consent before any “meaningful” action – that includes spending money, sending messages on your behalf, booking travel, or making other critical decisions. You can review what it’s about to do, pause or cancel running tasks, and you get a report once something completes, so you’re not left guessing what it changed. This lines up with how Microsoft has handled Copilot actions in Edge, where there are tiers of confirmation and stricter modes that always ask before acting on certain sites.

The positioning is also deliberately broad: Copilot Tasks is “for everyone,” not just developers or enterprises with complex workflows. That’s a subtle but important shift from the earlier wave of automation tools that assumed you’d wire up APIs, webhooks, and connectors yourself. Here, the promise is that you don’t need to configure agents or fiddle with runtime plug‑ins; you just describe the outcome you care about in normal language and let Copilot do the planning. If Microsoft can actually deliver that level of abstraction, it turns automation into something your parents could realistically use, not just a power‑user toy.

Of course, this is still early days. Copilot Tasks is limited to a waitlisted research preview, meaning most people won’t touch it for a while, and the initial feature set is likely to be constrained while Microsoft measures failure modes and abuse cases. But directionally, this is a clear signal of where productivity tools are headed: away from “AI as a smarter search box” and toward AI that actually logs in, clicks around, and gets things done while you do something else.

If Copilot Tasks works as advertised, it quietly redefines what “using a computer” looks like. Instead of you being the one stitching together email, calendars, forms, and web pages, you increasingly become the person who sets goals, checks summaries, and approves actions. In other words, you stop being the task‑runner and start being the editor‑in‑chief of your own digital life – with an AI agent in the background doing the legwork.