MICROSOFT 365 COPILOT KNOWLEDGE AGENT - INTRO
- Jonathan Stuckey

- Oct 15
- 6 min read
author: Jonathan Stuckey
This is the first of two articles about the Microsoft 365 Copilot Knowledge Agent. In this article I cover off what it is, how to set it up, and where the limitations in deployment are. In part 2, I'll outline what I found were best use-cases and scenarios and if it will cut-it for businesses today.
Microsoft has (soft) launched its latest in Agentic development releases under the "Copilot" branding as Preview. So lets de-construct what the Knowledge Agent is, what it does, its constraints, issues in Preview and where its going.
What is the Knowledge Agent?
Microsoft pitches the Knowledge Agent as "It enriches, organizes, and maintains SharePoint content in a structured, authoritative format—optimized for Microsoft 365 Copilot agents."
In short its a helper bot and a means to speed-up preparation of your documents and content held in SharePoint Online ready for use by AI. It does it by ingesting documents in your library, analysing the content and recommending & injecting specific metadata for use in sorting, filtering and finding (search) content - and to a limited extent - starting to apply some process automation

Its based on a combination of technologies strip-mined from Syntex content processing, SharePoint search index management, the Copilot LLM and Automate. It offers a smart approach to the age-old issue of addressing your old content which you never got around to sorting, categorising, labelling and (ultimately) cleaning-up.
So on the surface its value-proposition is:
it will identify your content,
sort and categorise these items, and
clean-up enough to allow you to do more with it (using Generative AI)
...all without hiring an expensive "body" in the form of an information manager or an analyst.
Basically its like having your mum sort your room out, enough that it looks better but (hopefully) not too much that you can't find anything afterwards.
How do we get hold of it to test?
IMPORTANT: THE KNOWLEDGE AGENT IS PREVIEW. IT HAS SIGNIFICANT LIMITATIONS TODAY
Well it is in Preview at the moment, and you must have bought Microsoft 365 Copilot licenses and allocated them to users. On the plus-side you only need 1 license to actually set it up!
Fortunately this isn't a limited access preview, or private sign-up. Microsoft wants a lot of people testing this agent because the more people use it, the more organisations will start to address the preparation necessary to support adopting Generative AI (read: paying for Copilot).
There's a whole load of guidance online, and unusually for Microsoft its mostly readable and seems to be accurate-ish - based on the Agents limited ability to execute at the moment.
How to get going
Before we start, you should know I do all my own stunts and this my work, including the pain of trying to get the flippin thing going in my tenancy. All the learnings below are because I'm very good at breaking things.
Pre-requisites:
You need to have bought and allocated a Microsoft 365 Copilot license,
be working with IT / someone with SharePoint Administrator role on your tenancy,
run PowerShell command prompt or ISE from a machine to connect to your tenancy,
can install version 16.0.26413.12010 (minimum) of SharePoint module (rel. 03/09/2025).
Enable your tenancy
There's no easy way to do this today, its PowerShell or nothing. So, launch your PowerShell console as Administrator...
check what module is installed and get the right version:
I could go through stuff about checking, updating etc, but the scorched-earth approach clears-out all potential for old versions so I suggest:
Get-Module -Name Microsoft.Online.SharePoint.PowerShell -ListAvailableUninstall-Module -Name Microsoft.Online.SharePoint.PowerShell -AllVersions -ForceInstall-Module -Name Microsoft.Online.SharePoint.PowerShell -RequiredVersion 16.0.26413.12010 -Forcepreset the powershell console to run what you need
Set-ExecutionPolicy -executionpolicy Unrestricted -scope CurrentUsersetup and connect to your tenancy via the SharePoint connect string
Connect-SPOService -Url https://organisation-admin.sharepoint.com -Interactiverun the (preview) Agent commands to enable tenancy sites
There are options, and the default without adding the -KnowledgeAgentScope is everything in SharePoint, the more appropriate way to do this is using:
Set-SPOTenant -KnowledgeAgentScope ExcludeSelectedSitesdefine the list of sites to Exclude from the use with the agent
CRITICAL: THIS IS AN EXPLICIT OPT-OUT MODEL WITH SITES.
You might not want to have test-users messing around with HR data, business contracts, investment strategy, executive meetings... documents ingested by the agent, so we'll identify all the sites to be excluded from the Knowledge Agents wide brief e.g.
Set-SPOTenant -KnowledgeAgentSelectedSitesList @("https://organisation.sharepoint.com/sites/administration","https://organisation.sharepoint.com/sites/StaffFiles","https://organisation.sharepoint.com/sites/Directors")...where 'organisation' is replaced by your tenancy name.
...and wait. At least 24-hours.
Troubleshooting
There were are number of specific issues ran into with enabling the tenancy for Preview, these related to trying to make use of the "Excluded" sites option. Most of the issues came with the following:
WARNING: The following URLs passed to KnowledgeAgentSelectedSitesList were invalid and ignored...
The reasons for those were:
So I've enabled it. Now what?
Well once you've had a goodnights sleep, and come back to the tenancy with a plan to test it (you have got a plan, right?) then your Copilot licensed user's will find...

Walking through the Agent's 'wizard' for primary tasks was straight-forward, if a very different experience to all the other Copilot and Agent options from Microsoft. If you want to see the basics up-close and personal then I'll be posting some videos shortly on my Youtube channel - once I've took out the ums and errs.
One the whole the end-user experience is pretty seamless, similar to working with the standard Microsoft Lists and SharePoint app templates walk-through. Basic. The disconcerting element was the potential to make an extraordinary mess of my library if you are not aware of the implications.
What are the limitations?
At the moment the limitations are numerous and (some) quite are a bit worrying, ranging from constraints on setup and enabling the Knowledge Agents scope of access, through to no controls on managing user-access with roles, or the limits on application when executing functions in your site libraries, generating new fields, creating views and labelling documents with metadata.
Some of the more interesting scenarios occur when you can run the Agent in live environment for the core scenarios of:
suggesting metadata for libraries based on the documents
creating new / additional views in the libraries, and
setting up workflows for "quick actions" on content
Suffice to say that the setup at the moment is "all-in", or explicit choice of opting out (some sites/everything in your tenancy),
Critical for me in assessing this were:
Microsoft recommend adding no more than 10 columns per library and processing files no larger than 65 pages.
If files from a folder are used to configure the library, the changes will be applied to the entire library.
The agent only processes the first 20 files during initial configuration. Additional files can be selected manually and processed... in batches of 20.
When managed metadata column types are used, it can only use the first 100 terms of a mapped term set. If a preferred term or synonym matches, the agent saves the preferred term to the column.
The pages at Get started with Knowledge Agent (preview) cover a lot of the restrictions, but the biggest one potential issues is the inability to ensure restricted or secured sites with custom permission-levels can be excluded.
What next?
While the above is a good starting point for enabling things - it's only that. In the next article I'll walk through just practical the suggested use-cases are, with the current strengths and weaknesses.
Resources
About the author: Jonathan Stuckey
Disclaimer:
Generative AI has been used in creating an image in the article, but only one! Everything else is real. All the irritation, griping and random comments are the responsibility of the author.





Comments