Microsoft has introduced a new class of AI assistants in SharePoint with SharePoint Agents. They access content from a site or library on a per-permission basis, and answer questions, summarize documents or support recurring tasks — directly inside SharePoint or Teams.

Important framing: An earlier version of this article reduced agent creation to "configured via Copilot Studio". In fact, Microsoft documents two distinct paths — see below.

What SharePoint Agents are

SharePoint Agents are AI assistants grounded on the content of a SharePoint site, a document library or a selected set of files. They respect each requesting user's existing permissions — what the user is not allowed to see, the agent will not answer either.

Typical tasks:

• Answer questions about a site's content (policies, FAQs, project documents)
• Summarize long documents
• Compare list or library content
• Return results with source citations in the chat pane

The two creation paths

Microsoft Learn describes two clearly distinct routes:

1. SharePoint-native (the right fit for most use cases)

• Every modern SharePoint site has a ready-made site agent, available via the Copilot icon in the top right and grounded on the site by default.
• A "Create an agent" experience inside a site or document library lets you build your own agent — no separate tool, no code. The result is an .agent file stored in the site.
• Site owners can mark such agents as "approved" so they appear in the site picker.

This path is the right choice for the majority of use cases: a knowledge agent for an intranet, a team FAQ agent, a contract assistant for a library.

2. Copilot Studio (for more complex scenarios)

Copilot Studio is the right choice when you need to:

• Combine content across multiple sites or include connectors,
• Define your own topics, knowledge sources or custom actions,
• Wire the agent into Power Automate flows,
• Surface the agent outside SharePoint (e.g. as a standalone bot in Teams or inside Microsoft 365 Copilot Chat).

Copilot Studio is more powerful, but it also requires a clear setup and the right licensing.

Realistic use cases for Swiss organizations

• Intranet knowledge agent: Answer HR, IT self-service or process questions — grounded on the official intranet site.
• Contract and proposal assistant: Find clauses, terms or counterparties across a contracts library.
• Compliance support: Check documents against internal standards (e.g. FADP / nDSG requirements).
• Project documentation: Summarize status reports and minutes from a project site.

Permissions, governance, security

• Access: Agents only see what the requesting user is allowed to see — site- and item-level permissions are strictly inherited.
• Sensitivity Labels: Confidential documents stay confidential; existing classifications are honored.
• DLP: Microsoft Purview Data Loss Prevention remains active.
• Restricted Access Control: Sensitive sites can be locked down specifically against Copilot/agent access.
• Audit trail: Actions are recorded in the Microsoft 365 audit log.

How we approach this

As a Microsoft Partner we typically start with Swiss organizations like this:

1. 1Check site hygiene: permissions, sensitivity labels, external sharing.
2. 2Pick a pilot use case: clearly bounded — an intranet FAQ is a good first step.
3. 3Try the SharePoint-native agent first: the built-in path covers most needs.
4. 4Move to Copilot Studio when justified: only when the use case truly goes beyond the native flow.
5. 5Support adoption: set realistic expectations, measure outcomes.

SharePoint Agents are not a silver bullet, but they are a practical way to make existing SharePoint knowledge more accessible — provided the underlying tenant is in good shape.

Further reading:

• SharePoint as an AI foundation
• Microsoft Learn — SharePoint Agents
• Microsoft Copilot Studio