Agents On by Default, AI Spend by Charge Code and Survey Results Worked Out in Code - PM Status Report, 9 November 2026
- What's New
- What It Means for Your Projects
- Where Things Stand
OpenAI made dots, its agents that keep working between conversations, generally available to ChatGPT Enterprise, Edu and Healthcare workspaces on 4 November. Two days later Anthropic let Claude Enterprise organisations tag projects with a charge code and see spend by tag. Microsoft had already begun limiting Copilot Chat for staff without a paid licence on 2 November.
What’s New
OpenAI made dots generally available to eligible ChatGPT Enterprise, Edu, and Healthcare workspaces, where they become available by default from 18 November unless an admin opts out. It also released GPT-6 Cyber to approved customers of its Daybreak Red security programme. OpenAI researchers used it to find 81 previously unknown high-severity flaws in Firefox, which Mozilla has begun fixing, and OpenAI rates it “Critical” for cybersecurity under its own risk framework.
Anthropic added project tags to Claude Enterprise in beta for eligible Enterprise organisations. An owner tags a Claude project with a charge code or client-matter number and can then see most spend in those projects by tag. Coverage of the beta says tags apply to chat, Cowork and Claude Code but not to API use. A day earlier Anthropic released a Security Center in beta, which compares an organisation’s Claude settings with Anthropic’s recommended configuration and exports the result as an audit report.
Microsoft began fair-use limits on Copilot Chat for users without a Microsoft 365 Copilot license in commercial tenants, from 2 November for organisations with 2,000 or fewer seats and from 9 November for larger ones. The limits can change through the day, Microsoft sets them, and admins can’t view or adjust them.
Google began rolling out actionable insights in Google Forms. Gemini analyses responses “by writing and running Python code” on the raw answers, finds relationships between questions and suggests next steps. It covers Business Standard and Plus, Enterprise Standard and Plus, Education Plus, and Google AI Pro and Ultra.
Elsewhere, Mistral expanded Mistral Studio with managed agents and workflows that set which steps follow fixed rules, where an agent reasons and when a person decides. Mistral’s own incident-investigation agent follows the team’s runbook and asks for approval before sensitive actions, and BNP Paribas runs Studio on its own infrastructure. Atlassian added Jira Align read and write tools to its Rovo MCP server, so a connected AI assistant can look up, create and update portfolio items.
In consumer and creative tools, ChatGPT Images 2.6 can turn a prompt or a sketch into a looping GIF and save an image by name to reuse in later chats. Google Vids generates clips of up to 15 seconds with Gemini Omni 1.2 Flash, keeping a product or character consistent from a reference image. Meta’s Muse added Spaces: describe a tracker, planner or project board in chat and Muse builds it, then keeps it current from connected apps. Anthropic opened a beta of Personal Finance in Claude for Pro, Max and Ultra users in the US, which reads linked bank and card accounts but can’t move money.
What It Means for Your Projects
Put the default-on date in the change calendar. When always-on agents become available to a whole workspace on a set date, a programme team can treat it like any system release: a line in the change plan, a note to the team on what arrives, and an agreed list of project folders the agents stay out of until a pilot has run.
Charge AI to the work package it served. On a capital project or a consulting engagement, tag each AI workspace with the code already used on timesheets, so the monthly cost report shows AI spend beside labour for each package and a client recharge matches the matter number.
Run the retrospective as a survey and bring the numbers into the room. After a system go-live at a hospital, a 15-question survey with 80 responses can show whether the teams that rated their training lowest also logged the most defects, with percentages calculated from the answers, which gives the retro its first question.
Hold a patch window in every release plan. On a software delivery or infrastructure upgrade, flaws found with AI help reach the project as batches of supplier patches, so reserve capacity in each release for them and name who decides whether a critical fix jumps the queue.
Write down which tasks run on the free assistant. If a business analyst drafts the requirements summary on an unlicensed assistant whose limits nobody on the project can see in advance, name that task in the plan with a licensed tool or a manual fallback against it.
Where Things Stand
Three of this week’s releases deal with who pays for AI and who switches it on. Agents available by default, spend sorted by charge code and free use with limits nobody can see all turn AI from a tool someone chose into a running cost and a team default. I think that puts AI where project managers already hold the controls, in the cost plan and the change plan, rather than in an IT budget the project never sees.
If your AI tools were charged to the project tomorrow, which work package would carry the biggest line?
Yes, AI helped me to write this :)