G'day,

Here’s something I believe many organisations are getting wrong about AI and SharePoint:

They think better AI means they no longer need to organise their content.

If Copilot can read, reason and search across our documents, why should we still care about metadata, content types and information architecture?

The answer became very clear during my recent conversation with Zach Rosenfield from Microsoft:

Zach has spent almost 20 years working across SharePoint and OneDrive. He helped build SharePoint Online into a platform supporting enormous global scale, and today his focus is AI for enterprise content.

We sat down to talk about Copilot in SharePoint, skills, SharePoint.md and where AI-assisted work is heading.

​Watch my full conversation with Zach →​

The 10,000-contract problem

Imagine you have 10,000 contracts sitting in a SharePoint library.

You ask Copilot to identify every contract containing a particular type of clause. If those contracts haven’t been classified, Copilot may need to inspect every document to provide a reliable answer.

That takes time, computing power and AI credits.

Now imagine those documents have already been enriched with useful metadata. Instead of reading 10,000 contracts, Copilot may be able to answer the question using a relatively simple filter.

The answer becomes faster, cheaper and potentially more reliable.

This doesn’t mean people should spend their days manually tagging documents. The real opportunity is to use AI to create and maintain more of that metadata for us.

But the metadata itself still matters.

That distinction is important because there’s a temptation to treat AI as a silver bullet—to point it at years of accumulated documents, inconsistent folders and abandoned sites and expect it to make sense of everything.

AI can help you clean up that environment, but it doesn’t magically remove the consequences of a messy content estate. The quality and structure of your information still shape the quality, speed and cost of the answers you receive.

Copilot in SharePoint is a “doer”

The bigger idea we explored is that Copilot in SharePoint isn’t simply another place to chat with AI.

Zach described Copilot Chat as more of a “talker,” while Copilot in SharePoint is designed to be a “doer.”

Copilot Chat is incredibly useful for asking questions, developing ideas and creating content. Copilot in SharePoint, however, sits inside the environment where the work already lives.

It understands that you’re working with a particular site, page, document, list or library. It also has access to tools and capabilities that SharePoint has accumulated over decades.

You don’t necessarily need to know how to build a workflow or use every SharePoint feature. You can describe what you’re trying to accomplish and allow Copilot to help perform the work.

That starts to look less like a chatbot and more like an agent you can delegate work to.

Think of Copilot as a new employee

One of my favourite analogies from the conversation was Zach’s description of Copilot as a new employee arriving for their first day.

The employee may be highly capable and have access to plenty of tools, but they don’t yet understand their job. They don’t know how your organisation operates, where information belongs or which internal rules apply.

Before they can perform useful work, you need to give them context.

That’s where skills and SharePoint.md become particularly interesting.

A skill can teach Copilot how your organisation performs a particular task.

Not how any company might respond to an RFP—but how your company responds to one.

Not how contracts are generally reviewed—but which rules, approval stages and escalation points apply inside your organisation.

The goal isn’t to teach AI everything. It’s to teach it what it could not possibly know about your business.

That might include your internal terminology, required document structure, decision rules, governance requirements or definition of a successful result.

Giving a SharePoint site a memory

If a skill teaches Copilot how to perform a particular task, SharePoint.md provides the context it should understand whenever it works on a site.

Zach described this as the site’s “group memory.”

It might explain that your financial year ends in June, that reports must be saved in a particular location or that your team uses a specialised definition for a common business term.

These details may seem small, but they’re what separate a generic AI response from an outcome that fits the way your team actually works.

The distinction is relatively simple:

• If Copilot should always know something while working on the site, it may belong in SharePoint.md.
• If the information only applies when completing one particular task, it probably belongs in the relevant skill.

This gives organisations a practical way to provide AI with shared context without relying on every employee to write the perfect prompt.

Instead of teaching every person how to communicate with AI, we can begin teaching AI how the organisation already works.

​Watch Zach explain skills and SharePoint site memory →​

Where this becomes practical

When you put all of this together, SharePoint begins to look very different.

It’s no longer simply somewhere to store documents. It can become the place where your organisation’s content, context, process instructions and AI capabilities live together.

The most interesting opportunities may not begin with enormous transformation programs. They may begin with one painful process inside one team—a process that matters but has never justified purchasing another major software platform.

Think about RFP preparation, contract review, financial reconciliation, sales operations or live operational reporting.

These are processes where people often spend hours finding information, checking documents, compiling updates and chasing the next step.

An agent that understands your content, your internal rules and the required outcome could remove a meaningful amount of that work.

SharePoint at the heart of the AI era

SharePoint has supposedly been “dead” more than once.

Yet enterprise AI depends on enterprise knowledge—and for many organisations, an enormous amount of that knowledge already lives in SharePoint.

That puts SharePoint right at the heart of what comes next.

Not because it’s a shiny new AI tool, but because it can provide the content, context, memory and governance that allow AI to perform useful work inside a real organisation.

The organisations that benefit most won’t be the ones that simply add a chatbot to an untidy content estate. They’ll be the ones that deliberately curate their knowledge, improve their metadata and teach AI the practices that make their business unique.

So, if you’re thinking about introducing AI into your organisation, don’t begin by asking:

“Where can we add a chatbot?”

Ask this instead:

That question will take you somewhere far more valuable.

​Watch my full conversation with Zach on YouTube →​

And I’d genuinely like to hear what you think. Would you like Zach and me to make these conversations a regular thing?

Just hit reply and let me know.

Daniel

P.S. Keep an eye on your inbox for more tips and updates coming your way!