Five changes pulled from the Microsoft 365 official release communications feed. Each one is plain English, with a takeaway and a question worth thinking about. Roadmap IDs are given so you can check the source.

The Facilitator agent in Teams meetings now detects questions that go unanswered and offers to find a reply using web search. If someone asks a question and no one responds, Facilitator posts a prompt; a participant can accept and the agent shares what it finds. Roadmap ID 558341, GA expected July 2026.

A Project Manager agent is being added to Microsoft 365 Copilot to help plan, organise and track work. It starts with core task and plan management and will add more features over time. Microsoft moved it from preview and now expects general availability in June 2026. Roadmap ID 516576.

Copilot Notebooks gain Mind Maps: interactive visual artifacts that map the key topics, themes and relationships in a notebook. Users can open a node to read a summary or ask the notebook chat for more detail. Available across OneNote and the Microsoft 365 Copilot app. Roadmap ID 559029, GA May 2026.

Microsoft 365 Copilot will scope your chat history to the Copilot endpoint you are currently using, with an option to switch to an All chats view. The aim is to stop chats from one surface cluttering another. Roadmap ID 559601, GA expected June 2026.

Three changes are coming to AI Interpreter simultaneous mode in Teams. Interpreter audio and live captions now use the same language the user selects. Admins can fully disable voice simulation. Users get more platform voice options, with voices assigned dynamically to help tell speakers apart. Roadmap ID 562035, GA July 2026.

Three developments from the wider market in the past 48 hours. They share a theme with the Microsoft changes: agents that take actions, and the work of keeping them safe.

Salesforce introduced Agentforce Coworker, a beta feature that places an AI teammate inside searchable interfaces so it can retrieve CRM context and take actions in place. Salesforce also updated its Agentforce admin and certification materials. Published 22 May 2026.

Anthropic updated Claude Managed Agents with self-hosted sandboxes in public beta and a research preview called MCP tunnels. MCP tunnels let an agent call a company's internal MCP servers through an outbound-only encrypted gateway, reaching private tools without opening inbound access. Published 19 May 2026.

Cybersecurity agencies in the United States, United Kingdom and Australia issued a joint warning about the security risks of agentic AI, pointing to a widened attack surface as agents gain credentials and permissions to act for people. The warning lands as agent products move from pilots into daily use.

Read the eight items together and one line runs through them. Microsoft's Facilitator now answers questions in meetings, its Project Manager agent will draft and update plans, and Salesforce has put an agent inside the CRM that can take actions. At the same time, Anthropic is building tunnels so agents reach internal systems, and cyber agencies in three countries are warning that agent permissions and identity are now the soft spot. The pattern is not smarter chat. It is software that does things on your behalf, in the tools you already use.

Five Microsoft 365 changes, three AI developments, and one editorial line connecting them: agents are moving from assist to act. Use these cards to check the key terms.