Microsoft now wants Copilot to write your team's wiki. The new SharePoint knowledge base turns email, chats, meetings and files into a library of markdown pages that people can read and agents can run on. The idea is sound, the evidence is a preview and an internal test.
- What launched. On 14 October 2026 Microsoft announced a team knowledge base in SharePoint that Copilot and agents create, curate and reference, rolling out to Frontier preview tenants with usage-based billing.
- How it is built. Work IQ's compilation engine turns a source corpus (SharePoint, Outlook, Teams, OneDrive, Office files) into a corpus map of entities, topics and summaries, then writes interlinked markdown files that both people and agents read.
- What the evidence says. Microsoft's internal tests show agent answers up to 4x faster and about 30% more token-efficient; the supporting research used a synthetic set. No customer results exist yet.
- What to do. Treat it as a governed pilot: clean permissions first, name a human curator, cap metered spend and measure corrections per page rather than pages generated.
Every company has a knowledge problem that software has never solved: the answer to "how do we do this" lives in someone's inbox, a Teams thread from March and a deck nobody can find. Wikis were meant to fix it and failed because people do not maintain them. Enterprise search helped people find documents but did not tell them which document was current.
Microsoft's answer, announced by Jeff Teper, Executive Vice President for Microsoft Apps and Agents, is to stop asking people to write the pages. In his words, the technology finally exists to build a feature he has wanted for a long time. The post went up on 14 October 2026 and the LinkedIn version drew more than a thousand reactions in a day, which says something about how widely the pain is felt.
This review looks at what the SharePoint team knowledge base actually does, how the stack around it (Work IQ, Copilot in SharePoint, Autopilot and the new governance tools) fits together, what it costs to try, and how it compares with the search-first products that already own this category. As with every article in the Applied AI series, each product and use case gets a verdict of proven, emerging or hype.
Where AI sits in the team knowledge workflow
Each step of the knowledge lifecycle, what Microsoft's new stack now does with AI, and how mature it is.
What did Microsoft announce on 14 October 2026?
The announcement is a new object inside SharePoint: a knowledge base that any site can have, including the sites that sit behind Teams channels and Microsoft 365 Groups. Team members contribute content explicitly: an email thread, a chat, a meeting recording, a folder of documents already on the site. Copilot then organises that material into a set of pages that stay current as new content arrives.
Three surfaces can work with it from day one. Copilot in SharePoint, which Microsoft says is now generally available, creates and curates the base. The knowledge base is readable directly from Teams. And Autopilot, the long-running "digital teammate" Microsoft added to the new Copilot app on 25 September 2026, can create and use it, including when a user simply emails Autopilot a request.
Microsoft frames the problem as two-sided. Humans suffer from fragmented, outdated information; agents suffer from missing team context, so they answer generically or wrongly. A single curated layer is meant to serve both. The company gives three worked examples: a legal team pooling context for contracts, an engineering team handing project context to coding agents and onboarding new engineers, and a sales team keeping account history so an agent can draft next steps or an account snapshot.
How does Work IQ turn raw content into a knowledge base?
The architecture diagram Teper shared is the most useful part of the announcement, because it shows this is not another chat box over a document library. There are three stages.
The first is the source corpus: SharePoint, Outlook, Teams, OneDrive, Word and PowerPoint content that members point at the base. The second is Work IQ's compilation engine, which runs offline over that corpus and builds a corpus map: a table of contents, extracted entities and an entity registry, topics and summaries. Microsoft describes this as ongoing enrichment, so the map is recomputed as the corpus changes rather than built once.
The third stage is the knowledge base itself: a library of interlinked markdown files such as a project overview, team processes, product specs, customer insights, meeting notes and a roadmap. The files sit behind a human-friendly interface for people and behind tools and skills for agents. New content is ingested in real time, and Copilot and agents query the compiled layer rather than crawling the original files.
That last point is the technical bet. Retrieval over raw documents is expensive and noisy; retrieval over a pre-digested, cross-linked summary layer should be faster and cheaper. Microsoft's October 2026 internal tests reported agent responses up to four times faster and roughly 30% more token-efficient when content was retrieved through the knowledge base. A research paper the company cites (arXiv 2609.37226) reports better cost efficiency and answer quality on multi-document questions than agents working from original files or other enrichment methods. The paper's test environment was synthetic, which is worth remembering before repeating the headline numbers.
Which products make up Microsoft's team knowledge stack, and how mature are they?
The cards below describe each product as announced or shipped on 15 October 2026: what the AI does, how it is built, what it costs where a price or billing model is published, and the best evidence we found that it works. Two non-Microsoft products are included as the benchmark the new feature is measured against; their pricing was not re-checked for this article. No vendor paid to appear and the order is not a ranking.
SharePoint team knowledge base
- What
- Copilot and agents build, curate and query a team knowledge base inside any SharePoint site, including sites behind Teams channels and Microsoft 365 Groups. Members push in email, chats, meetings and existing site content.
- How
- A library of interlinked, agent-friendly markdown files compiled by Work IQ's knowledge compilation engine. Content is ingested in real time and re-enriched on an ongoing basis; people read it through a human-friendly view, agents through tools and skills.
- Price
- Frontier preview tenants only; users must be enabled for usage-based billing, managed in the Microsoft 365 admin center. Microsoft says billing terms may change before general availability (checked 15 Oct 2026).
- Proof
- Internal Microsoft tests (Oct 2026) report agent responses up to 4x faster and about 30% more token-efficient; a linked research paper used a synthetic test set. No customer results yet.
Copilot in SharePoint
- What
- Chat over a site's documents, pages and lists; creates and curates the new knowledge base; answers questions grounded in site content.
- How
- Runs inside SharePoint with the user's existing permissions and Work IQ context; Microsoft says it is now generally available.
- Price
- Included with Microsoft 365 Copilot and SharePoint AI licensing; knowledge-base curation falls under usage-based billing (checked 15 Oct 2026).
- Proof
- General availability announced in the same post; Microsoft publishes no adoption figures for the SharePoint surface specifically.
Autopilot digital teammate
- What
- A long-running agent in the new Copilot app that can create and use the team knowledge base, take tasks over email and act on the team's behalf.
- How
- Lives in the Copilot app announced alongside Home and Code; reads the knowledge base through Work IQ tools and can be reached by emailing it.
- Price
- Requires Autopilot access, which Microsoft says rolls out on a separate schedule; metered under usage-based billing (checked 15 Oct 2026).
- Proof
- Announced 25 September 2026; no published customer outcomes.
Work IQ with SharePoint tools
- What
- The context layer that compiles the knowledge base and exposes SharePoint tools so agents, and soon coding agents and CLIs, can read a team's content.
- How
- Builds a corpus map: table of contents, entity registry, topics and summaries computed offline, then refreshed as new content arrives.
- Price
- Not priced separately; consumed through Copilot licences and usage-based billing (checked 15 Oct 2026).
- Proof
- Microsoft says SharePoint tools in Work IQ are now available in the Copilot app; CLI and coding-agent integration is roadmap.
SharePoint Admin Agent and governance
- What
- Natural-language tenant analysis and guided fixes across permissions, content lifecycle, storage and settings; reports on which agents reach SharePoint and OneDrive content.
- How
- Pairs the admin agent with catalog management (content clusters for bulk attestations and access reviews), a file-level EEEU permissions report and a Governance Hub for site owners.
- Price
- Part of SharePoint Advanced Management and Copilot admin tooling; some features such as cross-geo move monitoring are in preview (checked 15 Oct 2026).
- Proof
- Feature list is published; no independent audit of how well the agent-access insights catch oversharing.
Copilot Studio agents on the knowledge base
- What
- Custom agents built in Copilot Studio that read the team knowledge base, plus agent flows that keep it curated automatically.
- How
- Listed on Microsoft's roadmap with Cowork and Copilot in Teams integration and deeper Microsoft 365 Groups support.
- Price
- Not yet available; Copilot Studio itself is sold by message packs and pay-as-you-go (checked 15 Oct 2026).
- Proof
- None. The announcement lists it as a future direction with no date.
Glean enterprise search and agents
- What
- Permission-aware search and chat across 100+ workplace apps, with a knowledge graph of people, documents and topics that grounds its assistant and agents.
- How
- Indexes connected apps rather than compiling a new corpus; answers cite the source documents and respect each app's access rules.
- Price
- Quote-based per-seat subscription; not re-checked for this article.
- Proof
- Long list of named enterprise customers and years of production use; the strongest evidence in this category comes from search, not from auto-written knowledge.
NotebookLM and Gemini in Workspace
- What
- Source-grounded notebooks that summarise, answer and generate briefings from a chosen set of documents; Gemini in Docs, Drive and Gmail answers over Workspace content.
- How
- Users pick the sources; the model only answers from them and cites them. Team notebooks are shared like any Drive object.
- Price
- Included in Workspace Business and Enterprise editions with Gemini; not re-checked for this article.
- Proof
- Widely used in business and education since 2024; the source-grounded pattern is the closest existing analogue to a curated team corpus.
What does it cost, and who can switch it on?
There is no list price. Access depends on two gates. The tenant must be in Microsoft's Frontier preview programme, and each user must be enabled for usage-based billing, which administrators manage in the Microsoft 365 admin center alongside spending limits. Microsoft says plainly that billing terms may change before general availability, and that Autopilot access, which the knowledge base relies on for its "teammate" scenarios, rolls out on a separate schedule.
For finance teams this matters more than the feature list. Microsoft has spent 2026 moving advanced Copilot capabilities from the flat per-seat licence toward metered consumption. Curating a knowledge base means repeated compilation runs over a growing corpus, and every agent query against it is a billable event. During a preview the numbers will be small; the question to answer before a wide rollout is cost per answered question, measured on your own corpus.
How does it compare with Glean, Google and the search-first tools?
The companies that already sell "find what the company knows" have taken a different route. Glean indexes a hundred-plus applications, builds a graph of people, documents and topics, and answers with citations while respecting each source's permissions. Google's NotebookLM asks users to choose their sources and then answers only from those, with citations, a pattern that has spread quickly in businesses and classrooms since 2024. Atlassian's Rovo and Notion's AI do versions of the same thing inside their own suites.
Microsoft's approach is narrower and deeper. It covers one team's content inside Microsoft 365 rather than every app, but it compiles a new, maintained layer rather than indexing what exists. The trade-off is clear. Search tools are proven at retrieval and need no curation, but they cannot tell an agent which of three conflicting documents is the team's current position. A compiled knowledge base can, if the compilation is right and a human checks it. Companies that are not standardised on Microsoft 365 get nothing from this launch, and companies that are will probably run both approaches for a while.
Which AI use cases in knowledge management are proven, and which are hype?
A wiki nobody writes is still a wiki somebody must own
The failure mode of every knowledge system is not creation but maintenance. Microsoft's design helps with the first half: Copilot drafts pages and re-ingests new content in real time. The second half, deciding that a compiled summary is wrong, that a process changed, or that a customer insight should never have been in a shared base, is still a person's job, and automated curation flows are explicitly on the roadmap rather than in the product.
The evidence is also entirely Microsoft's. The 4x and 30% figures come from internal testing, the research paper ran on synthetic data, and there are no customer references because the feature is days old. None of that makes the idea wrong; it makes it an emerging feature that deserves a controlled pilot, not a line in next year's operating model.
How should a company pilot the SharePoint team knowledge base?
- Pick one team with a real knowledge problemChoose a team that onboards people often or answers the same questions weekly: legal intake, a product squad, a key-account team. Baseline how long it takes a new joiner or an agent to answer ten standard questions.
- Decide what feeds the baseThe knowledge base is only as good as what members push in. Agree which channels, mailboxes and site libraries count as sources, and keep HR, legal privilege and customer-confidential material out unless a site is scoped for it.
- Assign a curator before the preview startsCopilot drafts the pages; a named person reviews them for the first month. Track corrections per page as the quality metric, not page count.
- Run the governance reports firstUse the file-level EEEU permissions report and catalog management to fix oversharing on the chosen sites before any agent compiles them. A knowledge base inherits every permission mistake underneath it.
- Cap and watch usage-based spendCuration, Autopilot and agent queries are metered. Set spending limits in the Microsoft 365 admin center, review consumption weekly during the preview and model the cost per answered question before expanding.
💡 Pro tip: Ask ten questions a new joiner would ask on day one, run them through Copilot before and after the knowledge base is compiled, and keep the transcripts. That before-and-after set is the only evidence you will have when someone asks whether the metered spend was worth it.
What can software companies learn from how Microsoft built this?
The series looks at AI from both sides of the contract. If you build software that grounds assistants or agents in customer data, these are the product lessons in Microsoft's design.
Microsoft's design computes a table of contents, entities, topics and summaries offline, then serves agents from that map. If your product grounds agents in documents, a compiled layer is the lever for speed, cost and consistency.
Plain markdown files with links are the common denominator between humans, coding agents and CLIs. A format only your UI can render will lose to one every tool can read.
Agent-access reports, permission clean-up and attestations were announced in the same post as the knowledge base. Buyers in regulated industries will not pilot auto-written knowledge without them.
Microsoft calls this Frontier preview, says billing may change and labels its results as internal tests. That disclosure buys more trust than a launch-day case study would.
What are the risks of AI-curated team knowledge?
⚠️ Before you switch it on
Oversharing at scale. A knowledge base inherits the permissions of its sources. If a site is over-shared today, Copilot will compile confidential threads into neatly linked pages that more people and more agents can read. Run the permissions and attestation reports first.
Confident, wrong summaries. Compiled pages read as authoritative. A summary that merges an old policy with a new one, or attributes a decision to the wrong meeting, will be trusted precisely because it looks curated.
Unpredictable metered spend. Compilation, re-enrichment, Autopilot runs and agent queries are all usage-billed, and Microsoft says the terms may change. Set admin-center limits before the pilot, not after the first invoice.
Agent accountability. Agents reading and writing team memory need an audit trail. Use the agent-access insights and keep humans on anything the base feeds into contracts, HR decisions or customer commitments. Our guide to AI and the law covers the disclosure duties that apply from 2026.
Frequently asked questions
What is the SharePoint team knowledge base?
It is a library of interlinked markdown files that Copilot and agents create and maintain inside a SharePoint site, built from content team members contribute from email, chat, meetings and existing sites. People read it as pages; agents use it as context. Microsoft announced it on 14 October 2026 for Frontier preview tenants.
Who can use it today?
Only tenants in Microsoft's Frontier preview, and only users enabled for usage-based billing by their administrator. Autopilot integration requires separate Autopilot access, which is rolling out on its own schedule.
Does it replace Glean, Notion or a wiki?
Not yet. Search tools such as Glean index what already exists across many apps; the knowledge base compiles a new, curated layer for one team inside Microsoft 365. Most companies will run both for a while, and anyone outside Microsoft 365 is out of scope.
What does it cost?
Microsoft has not published a price. Access is gated by usage-based billing, meaning curation and agent queries are metered, and the company says terms may change before general availability. Budget for a pilot, not a licence.
Sources
- Microsoft Tech Community — Introducing a team knowledge base, curated by Copilot and agents (14 Oct 2026)
- Jeff Teper, LinkedIn — Announcing today: a team knowledge base, curated by Copilot and agents, powered by SharePoint
- arXiv — Research paper on compiled knowledge for multi-document agent questions (2609.37226), cited by Microsoft
- Future Work blog — What is new: a knowledge base for the team, kept by Copilot (15 Oct 2026)
- Microsoft — Bringing CRM into the flow of work and agents into business process (Jeff Teper, 8 Oct 2026)
- Vlad Talks Tech — Microsoft 365 Conference 2026 recap: SharePoint Skills, Agent 365
How we judge AI in software
Proven means the feature is generally available, used by a broad customer base and backed by measurable results we could trace. Emerging means it ships but adoption or evidence is thin, mixed or only vendor-reported. Hype means the announcement, demo or marketing claim runs ahead of what customers can verify today.
Every Microsoft product, availability statement and figure was checked against the company's announcement and reputable reporting on 15 October 2026; vendor-reported numbers are labelled as such. Non-Microsoft products appear as benchmarks and their pricing was not re-verified for this article. No company paid for inclusion or position, and nothing here is a purchase recommendation. Product names, availability and billing in AI software change quickly, so confirm current terms with the vendor before you buy.
Discover more from Kurums | Business Intelligence
Subscribe to get the latest posts sent to your email.