Data teams could limit what Copilot interpreted

The Power BI update on 15 July changed how the standalone Copilot experience found answers. It would first examine a curated report and then the semantic model. Data owners could limit search to prepared content, while verified answers connected common questions with a human-curated visual. Quality became a visible editorial responsibility rather than an output delegated entirely to the language model.

AI Data Schema gained field search and indicators for hidden fields. This helped model owners select the portion of a model suitable for AI. In project governance, approved measures for schedule and cost could be exposed while provisional calculations or personal data remained outside the schema. A verified answer still needed an owner and a date because a correct response could become stale when its model changed.

The notebook gathered the working evidence

Microsoft's July update for Microsoft 365 Copilot described notebooks supporting up to 100 references. New functions would produce briefs and frequently asked questions from the material, with mobile support also arriving. A project team could use this bounded space for minutes, requirements, plans and decision evidence. Its value came from a visible source set that collaborators could inspect, removing old documents and checking what a summary relied on.

In the same month, the admin centre gained new reports and agent controls. Reporting was intended to distinguish active agent users and agents by publisher. A project office could therefore separate a personal agent, an organisation-built agent and one supplied by a vendor. These categories did not prove safety, but they created a starting point for ownership, lifecycle and cost management.

Gartner's June poll suggested agent use was still early: 24 per cent of 147 IT leaders had a few agents and four per cent more than a dozen. Gartner used the findings to argue for monitoring controls. A simpler first step for project work was a register recording each agent, owner, data sources, permitted actions and stop conditions, linked to the usage reports.

A decision needed both evidence and an owner

A practical workflow could begin in a notebook, use Power BI for a controlled measure and end with a named decision in the project log. When Copilot proposed an answer, the recipient should be able to open the verified visual and inspect its filter and refresh time. The team could measure corrected answers, ageing references and decisions that cited approved evidence. AI support would then become a traceable part of collaboration.

Sources

Microsoft Power BI Blog: Power BI July 2025 Feature Summary, 15 July 2025.

Microsoft 365 Copilot Blog: What’s new in Microsoft 365 Copilot – July 2025, 31 July 2025.

Gartner: Guardian agents and controls for agentic AI, 11 June 2025.

Microsoft: 2025 Work Trend Index Annual Report, 23 April 2025.

For discussion

Which project questions need a verified answer, and who approves it when the model or figures change?