Blog · Data and analytics

AI agents now work inside Power BI and Fabric. They all read the same thing.

Microsoft's September and October 2026 Power BI and Fabric updates moved AI agents from demos into everyday tools. Four changes matter most for a mid-size company.

A team reviewing an analytics dashboard with charts and a breakdown graph on screen

Microsoft's September and October 2026 Power BI and Fabric updates moved AI agents from demos into everyday tools. Four changes matter most for a mid-size company.

Agents can build reports. Microsoft lists Power BI agentic experiences as generally available. Through skills and the Power BI Authoring MCP server, an agent can read and change semantic models and reports, and BI developers review and refine the result instead of starting from scratch. In October, the report authoring skill was extended to custom visuals.

People can ask Power BI from Microsoft 365 Copilot Chat. Power BI data answers in Copilot Chat are generally available. Users can attach a report or a link and ask in plain language. The answers use the data, measures, relationships and business logic already defined in Power BI, and existing Power BI permissions still apply.

Fabric data agents work as tools in Copilot Studio. This integration is generally available. A Copilot Studio agent can call a Fabric data agent next to its other tools, and the agent can be published to Microsoft Teams and Microsoft 365 Copilot. One setting deserves attention: with User authentication, each person needs access to the data agent and its data. With Maker authentication, everyone who uses the agent sees data through the access of the person who built it.

Model owners get a new switch. People with write permission on a semantic model can block read-only users from reaching it through Copilot, data agents and Microsoft's MCP tools. By default, access stays on.

What the three agents have in common: every answer comes from your semantic model. If two reports define revenue differently, the agents will repeat the disagreement, quickly and in more places.

A practical order of work:

  1. List the numbers leaders argue about most, often revenue, margin or headcount.
  2. Agree each definition in writing.
  3. Fix or build the semantic model so it returns those definitions, and certify it.
  4. Review the Copilot access setting on each model, and choose User or Maker authentication on purpose.
  5. Then connect the agents, and compare their answers with the certified report.

Talk to an Expert

Tell us what you're working on, and we'll tell you how we would approach it.

Talk to an Expert