Microsoft 365 edition
Copilot Cowork gets a slider between fast and thorough, Microsoft writes new rules for AI that acts on its own, and AI leaders are hitting the brakes themselves.
In Copilot Cowork, you now choose: a fast answer or a thorough answer
A new slider now decides how much thinking Copilot puts into your question. 'Scheduled' is now called 'Automations'.
Microsoft 365 Copilot Cowork got a big update this month. The most visible change: a new slider. It lets you choose, per question, whether you want a fast and cheap answer or a slower and more thorough one. Until now, Copilot decided that itself. The navigation on the left has changed too. Where it used to say 'Scheduled', it now says 'Automations'. That's where you set up recurring tasks. The choice between the GPT and Claude versions has moved to the prompt box itself, the place where you type your question. Cowork can also do more than before. It can now prepare full briefing documents, presentations and Excel overviews in one go, for example for a client meeting. So you're no longer asking for one piece of text, but for a set of pieces that belong together. Cowork is part of your employer's paid Microsoft 365 Copilot environment, not the free Copilot Chat. Whether you see it depends on what your admin has switched on. Microsoft often rolls out changes like this in stages, so your screen may still look different. This week, check once in Cowork whether the slider is there yet, and for a long task, set it to thorough on purpose. The slider controls how much computing work the model puts into your question. On 'fast', it answers right away based on what's obvious. On 'thorough', it takes more steps in a row: first figuring things out, then combining, then writing. That takes more time and computing power, so it costs more per task. The fact that you now see the model choice in the prompt box works the same way: you pick which model works on each question, instead of one setting for everything. A button you can adjust is also a button you can set wrong. Setting everything to 'thorough' takes time and uses up your organisation's Copilot budget faster. Setting everything to 'fast' gives you smooth text on complex pieces that just misses the point. The rule of thumb: short summary or standard email on fast, a piece going out to others or with figures in it on thorough. And keep checking the result. A more thorough answer isn't the same as a checked answer.
Background for subscribersFrom now on, you decide per question in Copilot Cowork whether the answer should be fast or thorough, instead of Copilot deciding that itself. Use fast for short jobs like a summary or standard email, and thorough for pieces going out to others or with figures in them. Also note the new name 'Automations' for what used to be called 'Scheduled', and the model choice that now sits right by your question. Always keep checking the result: a thorough answer isn't automatically a checked answer.
Microsoft writes new rules for AI that acts on its own
Microsoft has revised its own standard for responsible AI use. This is in the third version of the Responsible AI Transparency Report 2026, published on 8 September. New is a chapter for organisations that use AI, so also for employers using Microsoft 365 Copilot or GitHub Copilot. They must assess for themselves what can go wrong and arrange human oversight. For AI that carries out tasks on its own, extra requirements apply: strictly limit what such an agent can access, make sure a human can always step in, and monitor its behaviour through logs. The report includes real-world examples, including a phased rollout of Copilot with pilots and code reviews. Ask your admin who in your organisation is allowed to turn on agents and who keeps an eye on them. The principle is that an agent may only do what it needs to do. If it gets access to all of SharePoint, it can also draw on documents you wouldn't share. If you limit access to one folder or one team, it stays limited to what's there. The logs show afterwards which steps the agent took, so you can trace a mistake instead of only seeing the result. Human oversight sounds obvious, but in practice it fades. If an agent works well nine times out of ten, no one checks the tenth time anymore. So agree on which steps always need a human in between, for example before something goes out publicly or gets permanently deleted.
Background for subscribersThis is a taster.
Subscribers read the whole Microsoft 365 edition: every story, each with its background in plain language.
Subscribe →Not ready for that? Read the free letter first →
AI and tech, made simple.
This is the Microsoft 365 edition. Subscribers get the background to every story, in plain language.
Subscribe
inklaretaal · Amsterdam · © 2026