Copilot Cowork all you need to know

Episode one started in an Uber in Redmond. Episode two started after a holiday. Femke back from Lego Land in Denmark. Pascal somewhere between Swiss public holidays. Both back with fresh eyes — and a shared inbox full of things Cowork had already handled.
This is Frontier Insights, the LinkedIn Live series where two Microsoft MVPs share what they are actually seeing in the field. Not the polished keynote. The practitioner view.

What happened in the past few weeks?

Cowork has been in their hands for roughly a month. The experience has been equal parts impressive and educational. Femke opened one session with Cowork and asked it to send an email on her behalf. It sent the wrong one. “That was a lesson too,” she says. “You are not a passive observer. You are directing something that acts.”
That nuance matters. Cowork is not a chatbot. It is an agent that takes action in your Microsoft 365 environment. Which means the learning curve is not just about features it is about a new way of working.

Before you start: prerequisites

Not everyone has access yet, and that is intentional. A few things need to be in place:

Be enrolled in the Microsoft Frontier program via your Admin Center. Accept Anthropic as a sub-processor (Cowork runs exclusively on Anthropic models) and have an active M365 Copilot license.

One thing worth knowing: you do not have to roll this out to your entire organization at once. Assigning a focused group of AI champions is a smarter approach. Keep it contained, build real expertise, then expand.

Pascal also flagged that not everything works perfectly in this phase. Past chats occasionally disappear. Features shift between sessions. The UI changed during Femke’s vacation “agent mode” became “editing Copilot” overnight. This is frontier territory. Use the thumbs up and thumbs down buttons actively. Microsoft is listening.

We discussed a few use cases that you all can use. 
Use case 1: PowerPoint in your organization’s template
This was the use case Femke spent the most time on and the one that finally delivered on a promise the industry has been making for years.

She took an existing presentation on AI agents in the legal sector and asked Cowork to rebuild it in Wortell’s corporate template. It read the slides, generated graphics, applied the right colors and visual language, and returned a finished deck. “This is what I wanted two and a half years ago,” she says. “It actually works now. Maybe 90% ready. I tweak a few things. But I’m not starting from scratch anymore.”

Important distinction: this works in Cowork. It does not work the same way in the PowerPoint desktop app, and the browser-based PowerPoint agent asks a series of setup questions instead of just doing it. If you want the full template transformation, go through Cowork directly.
Pascal added a tip: Anthropic models have been producing noticeably better results for document creation in recent weeks than GPT models. That may shift as GPT-5.5 becomes more widely available within Copilot, but for now worth knowing.

Use case 2: Custom Skills
Pascal spent part of the weekend building custom skills. This is where Cowork starts to feel less like a tool and more like a configured colleague.
Skills are reusable instruction sets saved as markdown files in your OneDrive, under a dedicated Cowork folder. You create them with a slash command inside Cowork, describe what you want in plain language, and Cowork generates the skill file for you. No markdown knowledge required.

Pascal built a daily briefing skill. Every morning, Cowork sends him a Teams message with his priorities, meetings, and top tasks  automatically, using his calendar and inbox as the source. He also tested a project status report skill for when he is running a project as lead.
You can build up to 20 custom skills on top of Microsoft’s 13 pre-built ones. “Think about all the repetitive prompting we do every single day,” Pascal says. “Skills have the potential to absorb a lot of that.”
Once you have built a skill, you trigger it with /skills followed by the skill name. You can also copy skills between Cowork and M365 Copilot Chat  the markdown format is shared.

Use case 3: Computer use via the Researcher agent
This one generated the most surprise in the room.
Inside the Researcher agent in Cowork, there is a computer use capability. You give it a task, point it at a website, and it operates the interface for you moving the mouse, clicking, reading the screen, returning results.
Femke demonstrated it with LinkedIn. Every day she had been manually searching for new community members and sending each one a personalized direct message. With Cowork and computer use: fully automated. Right names, right messages. “I was flabbergasted. I’ve been doing this by hand every single day.”

The broader point is significant. Computer use works on any site where you have credentials  not just Microsoft surfaces. Internal SharePoint environments, paywalled publications, platforms without an API. If you can log in, Cowork can potentially act on your behalf. Pascal summarized it well: “You can still leverage the capabilities while acting as yourself, in environments that were never designed for automation.”

This is still a frontier feature. During the live session, Cowork briefly got blocked and then opened LinkedIn again independently. That kind of behavior is exactly why this phase is called testing but the concept is working.

Use case 4: Week planning and inbox management
Pascal closed with the use case most people will reach for first: organizing the week.

He used a pre-built Cowork prompt to review his calendar. Cowork flagged a public holiday with meetings still scheduled on it, identified a Thursday conflict, and asked which meeting to keep. Pascal answered. Cowork rescheduled everything, sent a decline email, and forwarded a full briefing to the colleague covering the delegated meeting without a single additional instruction.
“It knew the consequence of declining. That’s the difference between a chatbot and an agent.”

Femke tested inbox management separately. She asked Cowork to clean up her email. It produced a full overview of open messages with suggested next actions, deleted irrelevant mail after one confirmation, and drafted responses. She marks the outgoing drafts as “sent by Cowork”  a practice she is now applying to every Microsoft employee she emails. “It actually helps. That’s the most important thing.”

The bigger pattern

Across all four use cases, the same theme kept surfacing: Cowork is not doing clever things. It is doing the ordinary things  the daily admin, the repetitive tasks, the calendar conflicts so that you do not have to.
The challenge is not the technology. It is learning to direct it well. Femke’s misfired email is the clearest example. When you work with an agent that acts, the quality of your instructions matters in a different way than it does with a chatbot. You are not prompting for a response. You are delegating a task.
That is the mindset shift. And it is more significant than any individual feature.