What happened at Microsoft
Microsoft has had a run of outages across Outlook, Teams, Copilot and more than a dozen other services. Mark Gongloff wrote about it for Bloomberg Opinion, in a commentary CNA Business published on 16 November 2026. His point is that the outages are a bad look for the company after its AI-linked layoffs.
I couldn’t load the full article, so I’m working from the summary. I don’t know the dates, the length of each outage or how many jobs were cut. I also don’t know what caused the outages, and I won’t guess. What the commentary sets side by side is the layoffs and the outages. That is worth taking seriously on its own.
The claim behind AI layoffs
The claim behind this kind of layoff is simple. AI now does work that people used to do, so the company needs fewer people. For a lot of work, that is true. I think AI can already handle most of the transactional and operational work in customer service, for example. I’ve said before that those teams in their current form will shrink.
My first question is still whether there is a real need. What work did the AI take over, and who is left to check that it is done properly? Those are two separate lines in a budget. A company can cut the first and keep the second. When both go together, nobody is left to notice the work going wrong.
What changes and what doesn’t
AI changes how much work one person can direct. It lowers the cost of producing, drafting, counting and running routine steps.
It does not change who is responsible. You direct, the AI produces, you check and decide. You are responsible for every output that leaves your hands. A service that goes down is an output that left someone’s hands, and a customer cannot tell whether a person or a model made the mistake.
There is another point. Running a service takes people who know why each part exists, what depends on it and what to do when it breaks. That knowledge lives in the people who own the process. When they leave, the AI keeps producing, but nobody can tell good output from bad.
What an operator should take from this
Take a clinic with a booking system, or a shop with a WhatsApp inbox, or a marketing team publishing every week. Say you bring in AI to do the repetitive work. You will be tempted to count the savings as headcount.
I would count them as work removed. Then I would decide which people now spend their time on checking. Each process needs one named person who owns it, reads the output and can stop it. That person doesn’t need to do the work. They need to know what good looks like and have the authority to say no.
I judge an AI rollout by one thing. Does the service stay up, and do customers still get what they came for? Hours saved and tasks automated are only inputs. If the booking page goes down on a Monday morning, the savings figure on the slide means nothing.
Where this shows up in what I build
AeroChat is the clearest case in my own work. The business sets the limits on what the system can do, such as the refund amount or the discount. When a request goes beyond those limits, AeroChat passes the customer and the order details to your team. That design only works if a team is there to receive the handover, and I wouldn’t advise a merchant to remove it.
I use AI heavily in my own work. Claude does the building, the counting and the checking. I make the decisions. The checking is cheap now. Leaving it out would still be a mistake.
What to do
- List the processes your business runs on, such as bookings, orders, billing, replies and publishing.
- Name one person who owns each one. If the owner left tomorrow, ask whether anyone else could tell that the process had broken.
- Move AI into the repetitive steps first, and move people from doing those steps to reviewing them.
- Set rules in the system where a wrong result costs money or trust. Limits on refunds, prices and dates should be enforced by the platform, and not left to a prompt.
- Measure outcomes every week. Track uptime, response times and complaints. Treat a drop in any of them as a reason to look at the process before you look at the AI.
- Cut headcount only after a process has run well for long enough that you trust the checks. Then cut the work the AI took over, and keep the people who verify it.
Source: CNA Business: Commentary: Microsoft discovers AI can’t keep the email going