Model selection became dynamic in the workspace

Microsoft's August update said Microsoft 365 Copilot gained a GPT-5 button and real-time routing between a high-throughput model and one intended for deeper reasoning. A new Tools menu brought Pages, pinned agents, Researcher and Analyst into Copilot Chat. These functions rolled out in August, while some SharePoint-agent items in the same update were scheduled for September and did not represent August availability.

For a project team, routing made the processing behind an answer less visible. This increased the need to record the prompt, sources, time and intended use. A quick summary of meeting minutes and an analysis influencing budget or schedule needed different controls even when both began in the same chat.

Report context followed Copilot into SharePoint

The Power BI update on 12 August added Copilot to reports embedded in SharePoint Online. It also introduced preview answers and summaries that respected active report filters. A user could ask a question in the context of a bounded view. This helped collaboration, but recipients still needed to know the filter, refresh time and report version behind the answer.

Copilot for writing measure descriptions became generally available. A good description can help people and AI understand a figure, although an automated draft did not replace approval by the model owner. The definition needed to state calculation basis, period, unit and known limitations. Otherwise, fluent wording could create false confidence in an ambiguous measure.

On 5 August, Gartner placed AI agents and AI-ready data at the peak of inflated expectations. It stressed that data fitness had to be judged for the particular AI use case and recommended closely aligned pilots and cooperation between business and AI teams. This qualified the product updates: easier access to Copilot did not automatically make project data suitable for automated interpretation.

A decision card could preserve the trail

When a Copilot response informed a project decision, the team could retain a card containing the question, response, source links, Power BI report, filter, refresh time, model selection when visible and accountable approver. A monthly sample could be checked against the report and later outcomes. Error rates, missing sources and decisions based on stale data provided more useful governance measures than a count of Copilot conversations.

Sources

Microsoft 365 Copilot Blog: What’s new in Microsoft 365 Copilot – August 2025, 29 August 2025.

Microsoft Power BI Blog: Power BI August 2025 Feature Summary, 12 August 2025.

Gartner: Hype Cycle identifies top AI innovations in 2025, 5 August 2025.

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

For discussion

How should a project decision record its model, sources, report filter and accountable approver?