Article content
In briefShow moreShow lessCopilot Notebooks and Cowork gained more ways to share project context.
- Copilot Notebooks and Cowork gained more ways to share project context.
- Agent metrics and cost controls expanded in the admin centre.
- The project log needed to distinguish source, task and decision owners.
The working context followed the task
Microsoft’s July Copilot update described continued development of Notebooks, Cowork and agents in collaboration surfaces. Teams could gather documents, email and tasks and continue work without manually moving context. Automated actions could also turn provisional notes into real tasks more quickly.
Project governance therefore needed labels for reference, proposal and decision. A notebook could remain a living workspace, while baselines, budgets and accepted deliverables required versions. An answer needed to show the sources available to that user and their refresh time.
Administration created a measurable trail
June and July updates described cost and agent reports across groups and the organisation. They could connect consumption with a team and agent type, but activity did not prove correct analysis or useful work. The project office had to combine usage with errors, overrides and delivered results.
Gartner’s July customer survey showed people already used external AI tools to complete tasks. Project decisions could likewise rely on sources outside company control. A simple requirement for cited evidence and a named approver reduced the chance that an unverified response became a project assumption.
One log connected context with accountability
The project log could record the request, sources, agent, cost, created tasks and accountable approver. A weekly sample was checked against original data and plan status. Stale sources, duplicate tasks and ownerless decisions created concrete improvements for analysis, automation and collaboration.
Controls in practice
When a shared chat continued with another user, permissions and source sets could change. The project needed to show which parts were retained text and which were retrieved again. Confidential evidence could not travel through a sharing link to a recipient lacking the same access.
Cost data could reveal agent runs with repeated attempts or poor completion. Before setting a limit, the team needed to define completion: a summary, an approved decision or an executed task. Project leaders could then compare consumption with actual progress and improve both instructions and data access. The sample also needed abandoned runs, because completed tasks alone hid the cost of retries, missing access and manual repair.
Sources
Microsoft 365 Copilot Blog: What’s new in Microsoft 365 Copilot – July 2026, 31 July 2026.
Microsoft 365 Copilot Blog: Cost Management for Copilot, 16 July 2026.
Microsoft Dynamics 365 Blog: An agent-ready business applications platform, 23 July 2026.
Gartner: Customers favour third-party GenAI over company chatbots, 8 July 2026.
For discussion
Who owns the source, agent task and final decision in a shared workflow?









