I’ll be honest — I’ve had a rough few weeks. My site couldn’t handle the traffic it was getting, and I had to step back from publishing while we sorted it out. The irony is that the thing I most wanted to write about kept getting bigger and bigger while I waited. Because SharePoint hasn’t just released a new feature or two. It’s undergone what I’d call a genuine shift in how the platform works — and what it expects of us.

This post is going to be comprehensive. I want to cover everything I’ve tested, everything that’s been announced, and give you my honest take on what matters, what’s still maturing, and where I think this is all heading.

Let’s start with the thing I can’t stop talking about.

The SharePoint AI Agent: Building With Natural Language

If you haven’t seen this yet, it’s genuinely worth getting excited about. Microsoft has been rolling out what it calls “AI in SharePoint” — a natural language building experience that lets you describe what you want to create, and then SharePoint builds it. Not just creates a blank template. Actually builds it, with structure, columns, metadata, views, forms — the works.

I’ve been testing this with lists specifically, and I want to be straight with you: it’s one of the most significant time savers I’ve encountered in 25 years of working with this platform.

Here’s what the AI agent can do with lists right now: you can fully create, read, update, and manage list data through natural language — populate lists from files, evolve schemas over time, edit and refine items, format views, create forms, and ask questions grounded directly in your list data.

That last part is worth pausing on. You can ask questions about the data in your list, in plain English, and get answers back. For teams tracking projects, assets, requests, or anything else in Microsoft Lists, that’s genuinely useful.

It’s not just lists

The agent works across the core components within a SharePoint site. I want to be honest about something here though, because I think it’s important to set the right expectation: in my own testing, I’ve needed to create the site first and then work with the agent inside it. I couldn’t get site creation itself to work through natural language. That said, Microsoft has site creation via natural language on the roadmap — it was scheduled to begin rolling out through the preview from the end of March — so it may be available depending on your tenant settings and how far the rollout has reached. If you’ve had it work differently, I’d love to hear about it in the comments. This is a fast-moving preview and not everyone will have the same experience.

Document libraries are where I’ve had the most fun beyond lists. You describe how you want the library to work — the metadata columns you need, the views, how files should be organised — and the agent builds it. No clicking through column settings one by one. No manually setting up views. It just does it. SharePoint can also extract and apply metadata automatically as files land in the library, which means the structure stays maintained without someone having to tag everything manually.

And for pages, you can draft, refine, and reorganise content directly on the canvas. You can ask SharePoint to adjust tone, generate summaries, or reshape sections.

One thing that makes this different from other AI tools

A lot of AI tools give you a starting point. You still have to go in and clean it up. What I’ve found with the SharePoint agent is that it actually iterates with you inside the site — asking clarifying questions, proposing what it’s going to build before it builds it, and refining based on your feedback. That collaboration model is what makes this feel different from just “AI generates stuff.” It’s more like working with a skilled junior who wants to get it right before they start.

Converting Tables and Documents Into Lists

One of the practical things I keep coming back to is what this means for all the spreadsheet-based tracking that most organisations still rely on. Because there’s another angle to the list agent that’s genuinely useful: if you’re working from a document, you can convert a table to a list in one step, and it infers column types like date, currency, and person when it can.

Think about how many Excel files are floating around your organisation right now that are functioning as makeshift databases — project trackers, asset registers, contact lists, request forms. The list agent can take those and turn them into proper SharePoint lists, with validation, required fields, and views, without any manual column-by-column setup.

That’s a meaningful step toward getting information into governed, searchable, Copilot-ready structures — which matters more than ever right now.

SharePoint Skills: Teach SharePoint How Your Organisation Works

This is the piece I haven’t had time to fully test yet — but I’m genuinely excited about it, and I want to share what I know because I think it could be significant.

Skills were announced at the Microsoft 365 Community Conference in late April 2026, and they’re now rolling out to AI in SharePoint public preview tenants. The idea is straightforward but powerful: instead of AI in SharePoint being a generic assistant, you can teach it how your specific team or organisation works.

There are three things you can teach it:

• What to Know — Ask AI in SharePoint to remember something, and it will use a context file in the site to remember the norms and procedures for that site. So if your team always uses a particular naming convention, or has a rule about how projects are categorised, you can tell SharePoint and it will remember it going forward.
• How to Act — Ask AI in SharePoint to store a repeatable process and it will create a skill that can be reused by team members. Think of these like saved macros for AI tasks. A finance team could create a skill for generating quarterly reports. An HR team could create one for reviewing job applications against a template.
• What to Produce — Ask AI in SharePoint to create content and it will generate Word, PowerPoint, or Excel files, and can also generate custom reports and dashboards aligned to the content preferences of your organisation.

The technical side of how Skills work is interesting too. Custom skills enable users with edit permissions to create reusable, multi-step AI tasks using natural language, stored as Markdown files in the site’s Agent Assets library. This feature respects existing permissions and requires no tenant-level activation.

That means anyone with edit rights on a site can create a skill. It’s governed by the same SharePoint permissions and compliance policies that already apply to your content. No special setup, no developer required.

General availability for custom skills is expected to roll out between late May and early July 2026, so if you’re on the public preview now, you can start experimenting.

The SharePoint Admin Agent: Governance Gets an Upgrade

For admins and IT leads, there’s a separate but equally significant development: the SharePoint Admin Agent.

The SharePoint Admin Agent brings proactive, AI-driven governance into the SharePoint admin centre. It’s adding new agentic skills across a number of critical governance areas — including site permissions analysis, which uses natural language to analyse tenant-wide permissions, quickly flag oversharing risks, explain root causes, and provide clear remediation guidance.

It also covers site lifecycle management — surfacing inactive, ownerless, and high-risk sites with clear cleanup options — and storage management, giving admins a view of trends and risks across the tenant.

If you’ve ever tried to figure out where your SharePoint oversharing is happening using the old reports, you’ll appreciate how significant this is. What used to require PowerShell, data exports, and a fair amount of manual investigation can now be asked in plain English.

The Admin Agent requires a Microsoft 365 Copilot licence and is available now in the SharePoint Admin Centre.

The Bigger Picture: What Is SharePoint Becoming?

I’ve been working with SharePoint for 25 years. I’ve seen the platform go through a lot of changes. Some of them have been genuine improvements. Some have been shuffling deckchairs. This feels different.

What Microsoft is building is a platform where the structure of your SharePoint — your metadata, your site architecture, your library organisation — is the fuel for AI. Not just useful for humans navigating a filing system, but the actual data layer that Copilot and every AI-powered tool runs on.

That means the messy SharePoint environments that organisations have been tolerating for years aren’t just frustrating — they’re now genuinely limiting. Copilot can only be as good as the content it’s reasoning over. If that content is unstructured, poorly tagged, and dumped in random places, the AI results will reflect that.

The good news is that Microsoft is investing heavily in making it easier to get your SharePoint into shape. The SharePoint agent helps you create proper list and library structures from scratch. The Admin Agent flags the governance problems you didn’t know you had. And Skills are about to make it possible to bake your team’s specific processes directly into the platform.

The tools are becoming more capable. But they’re also making it clearer than ever that good structure is not optional. It’s the foundation everything else is built on.

What I’d Suggest Doing Right Now

• Opt into the AI in SharePoint public preview if you haven’t already. You don’t have to turn it on for all users — start with a pilot site and get a feel for what it can do.
• Try the list agent for something real. Pick a list you need to build and describe it in natural language. See what happens. I think you’ll be surprised.
• Check your metadata story. The less metadata your libraries have, the more you’re limiting what Copilot can do with your content. Even filling in basic columns makes a real difference.
• Watch the Skills rollout. This is the feature I’ll be testing hard over the next few weeks. If it does what I think it can do, it changes how we think about customising SharePoint for specific team workflows — without code.