I was at the Digital Workplace Conference in Melbourne last week and kept hearing people talk about Skills in Copilot Cowork.

It stopped me in my tracks.

Not because the concept was unfamiliar, but because I’ve spent the last couple of months exploring and building Skills in Claude. And suddenly the same word was being used by different people, in different conversations, to describe something that looked familiar on the surface but clearly wasn’t quite the same thing.

I came home a bit muddled, sat with it for a while, asked Loryan a few questions, and tried to work out where the overlap is real and where it’s just naming.

Here’s what I found.

First, some context

I work across both Microsoft Copilot and Claude. They’re both AI tools I use daily now, and both are things I help others navigate through my work.

Both platforms are using the word ‘Skills’ to describe a particular capability. Both mean something genuinely useful by it. But they are not the same thing, and confusing them (as I did, briefly, at that conference) will send you in the wrong direction.

What a Skill is in Copilot Cowork

Copilot Cowork is Microsoft’s move from AI-as-chat to AI-as-execution. Rather than answering your question and handing the work back to you, Cowork takes on multi-step tasks across Outlook, Teams, Word, Excel and SharePoint on your behalf.

A Skill in Cowork is a reusable set of instructions that tells Cowork how to handle a specific type of task. Instead of describing what you need from scratch every time, you capture the process once (such as how you want your meeting follow-ups written, what your weekly report should cover, what format a client brief should follow) and Cowork applies it automatically when it decides the Skill is relevant.

Technically, it’s a Markdown file (a SKILL.md) stored in a specific folder in your OneDrive. It has a name, a description, and plain-language instructions. Cowork reads your Skills at the start of each conversation and loads the ones it thinks apply.

It’s a sensible idea. The friction can sit in how it gets shared.

Each Skill lives in the individual user’s OneDrive. There’s no native way for an organisation to publish a Skill once and have it appear for every employee. If you want consistent Skills across a team, you’re currently looking at Power Automate flows to push the files out, or manual copying. It works, but it’s a layer of administration sitting between the person who built the Skill and the people who need it. Keep in mind this is the now… and things change constantly… so I may already be wrong!

There’s another thing worth knowing: Cowork Skills only apply when you’re working through the Cowork interface itself. Not when you open Word or PowerPoint directly and use Copilot from inside those applications. That’s a different surface, and it doesn’t read Skills from either platform.

What a Skill is in Claude

Skills in Claude are also reusable sets of instructions, but from my reading the architecture works quite differently.

Rather than sitting in individual users’ storage, Claude Skills are managed centrally. A skill author creates a Skill once, and it can become available to everyone in the organisation automatically. No file distribution. No per-user setup. The employee just opens Claude, starts a conversation, and Claude loads the relevant Skills in the background based on what’s being discussed. I have built Skills in Claude and shared with my team. I found it smooth, straight-forward, and easy to share for broader use.

An employee doesn’t need to know the Skill exists. They just ask Claude to do something, and the Skill quietly shapes how Claude responds, whether that’s following brand guidelines, applying a specific process, or drawing on particular organisational knowledge.

Because Skills are managed centrally, updating them is also clean. Change the Skill once, and every conversation picks up the new behaviour from that point on.

Where they’re genuinely similar

Both approaches are trying to solve the same underlying problem: AI tools are general by default, and general doesn’t serve most organisations very well.

Your company has specific ways of working. Specific standards. Specific language. A Skill is how you encode that specificity and make it consistently available to the AI.

The intent is the same. The file format is similar, both use plain Markdown with a name, a description and instructions. If you’ve built one, the structure of the other will feel familiar. Both also load Skills automatically rather than needing you to invoke them manually. You don’t say “use the brand guidelines Skill.” You just ask for the thing, and the right Skill applies itself.

Where they differ in practice

The main practical difference is distribution and scope.

	Claude Skills	Copilot Cowork Skills
Where they live	Centrally managed	Each user’s OneDrive
How they reach people	Automatically, for everyone	File distribution required
Updating them	Central, applies immediately	Per-user, or via automation
Where they apply	Any Claude conversation	Cowork interface only

Why the naming confusion matters

When I heard ‘Skills’ at that conference, my first instinct was that I already knew what was being discussed.

I didn’t.

I had to slow down and ask some fairly basic questions before I could follow the conversation properly. And I’ve been doing this stuff daily for months. Yet I felt I suddently wanted to dive into the M365 platform and try and see where it was showing up, how used and what it meant.

The risk is that people assume these are the same thing, and make decisions based on that assumption. An organisation thinking about whether Copilot Cowork Skills will solve their brand consistency problem might not realise the Skill only works in one specific interface, and only if each employee has the file sitting in their own OneDrive. That’s a different answer to “yes, we have Skills.”

A practical way to think about it

If someone mentions Skills in an AI context right now, the first question worth asking is: which platform, and where does the Skill live?

The concept is the same – a reusable set of instructions that shapes how an AI handles a task.
The implementation, the distribution model, and the surfaces where the Skill applies are all different depending on the platform.

Both are worth understanding. Both are genuinely useful.

They’re just not the same thing, and the shared name makes it easy to miss that.

I know, because I missed it.