AI in SharePoint (Formerly Knowledge Agent): Building Sites with Natural Language

March 17, 2026

6 mins read

Rollout status: Currently rolling out in Public Preview (March 2026) to Microsoft 365 Copilot–licensed users in opted-in tenants. Check your Microsoft 365 Message Center for availability.

⚠️ License required: Microsoft 365 Copilot add-on — AI in SharePoint is included with this license at no additional cost, but it is not available in standard Microsoft 365 plans. Check with your IT admin if you’re unsure what you have.

If you’ve been watching the SharePoint AI story unfold over the past year, you might be feeling a little déjà vu right now — in the best possible way.

What started as Knowledge Agent — a public preview launched in September 2025 — has grown, evolved, and just got a whole new identity. As of March 2026, it’s officially called AI in SharePoint, and it’s a significantly bigger deal than the name change suggests.

This isn’t just a rebrand. It’s a signal that Microsoft is treating AI as a native part of how SharePoint works — not a bolt-on feature, not a separate tool you open in another tab. AI is now baked directly into the platform, accessible right where you already work.

Here’s the full story: where it started, what it can do now, and why this matters for your organization.

So… What Was Knowledge Agent, Anyway?

When Microsoft introduced Knowledge Agent in September 2025, it was positioned as a major step toward making SharePoint content smarter. The core idea: AI shouldn’t just retrieve your documents — it should understand them, clean them up, enrich them with metadata, and answer questions grounded in what’s actually in your libraries.

In practice, Knowledge Agent could:

Auto-fill metadata across document libraries — tagging files by client, date, contract status, whatever mattered to your team
Generate AI-powered views that sorted and grouped content based on that enriched metadata (e.g., “policies expiring in 2026” or “contracts grouped by client”)
Answer questions about site content directly, with answers grounded in your actual documents — not a web search
Build automation rules from plain language (e.g., “email me when an invoice over $500 is added”) without any Power Automate expertise

For teams drowning in untagged documents and inconsistent libraries, this was genuinely useful. Over 5,000 tenants opted into the preview. Real organizations — Takeda, Mars, Hertz — started building on it.

But Knowledge Agent was always just the beginning.

What Changed: The Evolution to AI in SharePoint

Fast forward to March 2026, and Microsoft announced that all Knowledge Agent capabilities have been integrated natively into the platform — and the scope has expanded considerably.

AI in SharePoint now covers the full lifecycle of building and evolving SharePoint solutions: not just managing existing content, but creating it from scratch, in natural language, in a multi-turn collaborative conversation.

The tagline Microsoft is using captures it well: “You start with intent.”

Describe what you want to build. The AI generates a structured plan — spanning sites, pages, lists, and libraries — and executes it. You review, refine, iterate. No template hunting, no clicking through site creation wizards, no manual column configuration.

Think of it like having a very capable SharePoint consultant in your corner who can both plan and do the work — and who never sighs when you say “actually, can we add a column for status?”

Why This Is a Big Deal

Here’s what actually matters for your team:

You don’t need to know SharePoint to build in SharePoint. Natural language planning means non-admins and content owners can drive site creation without submitting IT tickets or taking a three-hour training. That’s a genuine shift.
It closes the gap between intent and execution. How many times has a SharePoint site been “good enough” because nobody had time to configure it properly? AI in SharePoint helps teams go from “we need a project hub” to a working, structured solution — fast.
Knowledge Agent’s best features didn’t go away — they got promoted. Metadata auto-fill, AI-generated views, and grounded Q&A are still there, and they’re now the foundation the new capabilities build on. The content quality improvements you get from metadata enrichment directly improve how well the AI can plan and build.
Custom skills let you encode your organization’s rules. You can define reusable “skills” — packages of business logic, terminology, and governance rules — that shape how AI behaves within your environment. These aren’t post-hoc guardrails; they run during execution. That’s a meaningfully different approach to governance.
SharePoint is now the top grounding source for Microsoft 365 Copilot. Microsoft has been explicit about this. The better your SharePoint content is structured and enriched, the better Copilot performs across your whole M365 environment — Teams, Word, Outlook, all of it. AI in SharePoint directly improves that foundation.

How to Get Started

📌 Admin note: AI in SharePoint is opt-in only. IT admins must enable it via PowerShell before users see anything. If you’re not seeing the floating action button, this is why.

Verify your license — Users need a Microsoft 365 Copilot license. No additional cost for AI in SharePoint, but the base Copilot license is required.
Opt in your tenant — Admins use the SharePoint Online Management Shell (version 16.0.26615.12013 or later) and run Set-SPOTenant with the KnowledgeAgentScope parameter. If your tenant was already opted into the Knowledge Agent preview, you’re already in — no action needed. (Note: The underlying PowerShell parameters still use “KnowledgeAgent” naming for compatibility during preview. That’s normal.)
Enable Anthropic as a sub-processor (optional but recommended) — AI in SharePoint’s refreshed preview uses an advanced reasoning model for multi-step planning. To get the full experience, admins should enable Anthropic as an AI sub-processor in the Microsoft Admin Center. Without it, the feature falls back to an alternate model and some capabilities may be limited. 📌 Admin note: This setting lives in the Microsoft 365 Admin Center under privacy/data processing settings.
Look for the floating action button — Once your tenant is opted in and you have a Copilot license, you’ll see a floating action button in SharePoint. That’s your entry point. It’s context-aware — it surfaces suggestions based on where you are in SharePoint and what content is nearby. [Screenshot: Floating action button appearing on a SharePoint site homepage]
Start with a natural language prompt — Try something like: “Create a project tracking site with a document library, a task list, and a status page for stakeholder updates.” The AI will generate a plan, walk you through it, and execute on your approval. [Screenshot: Multi-turn planning conversation in AI in SharePoint]
Refine and iterate — This is a conversation, not a one-shot generation. Push back, add requirements, change the structure. The multi-turn experience is designed for this.

Quick Tips

Start with a messy library. If you have a document library that’s never been properly tagged, AI in SharePoint’s metadata enrichment is a great first test. Point it at a library of contracts or HR docs and watch it auto-classify. The before/after is surprisingly satisfying.
Create skills for repeatable patterns. If your team builds similar project sites every quarter, encode that structure as a skill. From then on, anyone can trigger it with a single prompt — and it’ll follow your standards every time.
Don’t expect miracles on a poorly organized site. The AI reasons over your content, which means garbage in, garbage out still applies. A site with inconsistent naming, no metadata, and outdated pages will produce weaker answers and less useful structure suggestions. AI in SharePoint rewards investment in content quality.
Check the floating action button in different contexts. It behaves differently on a document library vs. a site homepage vs. a page. Exploring those context-specific suggestions is worth a few minutes.

Best Used For

Teams spinning up new SharePoint sites who want a structured starting point — project hubs, HR portals, department intranets
Document libraries with backlog content that needs metadata cleanup
Content owners who want their SharePoint sites to answer user questions without anyone manually maintaining a FAQ page
Organizations looking to improve Copilot answer quality across M365 by enriching their SharePoint content foundations

When NOT to Use It / Things to Watch Out For

Government tenants, you’re out for now. AI in SharePoint isn’t currently supported in Microsoft 365 Government (GCC, GCC High, DoD) or air-gapped environments. No timeline given yet.
It’s still preview. Some capabilities may not be available in all SharePoint surfaces — lists and SharePoint sites embedded in Teams have had inconsistent behavior. Test before rolling out broadly.
Multi-geo tenants have extra steps. If your tenant spans multiple geographies, the PowerShell opt-in must be run in each geo separately. Easy to miss.
The Anthropic sub-processor requirement will raise eyebrows in some organizations. For tenants with strict data residency requirements or procurement controls around AI vendors, this setting will need legal or compliance review before enabling. The fallback model works — it’s just less capable for complex multi-step planning.
It won’t save a neglected site. AI in SharePoint is an accelerant for teams already doing decent content hygiene. If the underlying content is a mess, start there first.

Wrapping It Up

The Knowledge Agent → AI in SharePoint arc is a good example of Microsoft actually following through on a preview feature. What started as intelligent content cleanup has grown into a full natural-language interface for building and evolving SharePoint solutions — and it’s now natively integrated, not a separate add-on.

For organizations that have been waiting for SharePoint to feel less like a filing cabinet and more like a working knowledge system, this is the clearest signal yet that the platform is heading somewhere genuinely different.

If you’re already opted into the Knowledge Agent preview, start exploring the new capabilities — especially the natural language site building. Pick one site your team has been meaning to improve, open the floating action button, and describe what you want. You might be surprised how far a conversation gets you.

And honestly? There’s something a little delightful about describing a SharePoint site out loud and watching it take shape. It shouldn’t work as well as it does — but it does.