On Friday 13 November, when Outlook was down, I had several chunky bits of work on the go – all cooking nicely in the background with various AI assistants. Despite the outage, they were all still there. I even made it into some calls because the Teams app on my phone had remembered the meeting codes. And for a few hours, nobody emailed me.
By Friday afternoon, emails had started dribbling in. By then, if I’m being entirely honest, I’d mentally checked out for the week and was slightly cross with Microsoft for getting it working again. Surely they could have held off until Monday. Three new episodes of Rivals had landed on Wednesday, so I imagine not everyone was quite as focused on restoring productivity as Microsoft.
What amazed me was how much I’d still been able to get done, despite the lack of emails and the lure of ’80s nostalgia. A few years ago, in my previous life as a civil servant, my inbox would have been where I was doing a huge amount of my work: checking room numbers for meetings, sending a draft, getting comments, asking a question, sending another draft. Without it, I’d have been missing both the conversation and whatever was attached to it. Even working on something offline would have been a gamble, because the discussion could have moved on or someone could have emailed a crucial bit of information.
Some of that back-and-forth now happens in Teams. Some happens with an AI assistant, working through an argument or improving a piece of writing. In my case, enough of the work had moved out of the inbox that losing it didn’t bring everything to a halt. It mostly removed the interruptions.
That was fortunate for me. But I know it wasn’t everyone’s experience.
What actually went wrong?
According to Microsoft’s preliminary incident report, Teams traffic had been moved onto the same set of gateway servers that handled connections to its email service. Moving traffic like this is something Microsoft does routinely to keep things running. This time, some of those servers were already out of service, and the extra load exposed a software bug that had been sitting there for months.
Think of a supermarket where a few tills are already closed. One of the remaining tills packs up, so everybody moves to the others. Except the extra customers make those tills pack up too. Even when you reopen one, it immediately gets the queue that’s been building up while it was closed.
That’s (very roughly) why sending people a different way, or simply restarting the servers, couldn’t get everything working again. Microsoft had to fix the bug and bring the servers back in stages. And we had to wait, with varying degrees of patience.
So the work has moved. Have the dependencies?
The tools I was using, including the Leading AI tools I had on the go, carried on working for the jobs I was doing. But Microsoft’s Copilot was affected, and Reuters reported problems with the Microsoft connectors used by OpenAI and Anthropic too. Microsoft said features that depended on its Exchange Online service were affected, while those that didn’t rely on it remained functional.
So “is the AI working?” turns out not to be quite the right question. The assistant might still have been there, eager as ever to help, but unable to get to the information you want it to use.
There’s a difference between asking the AI to help unpack an argument in a document you’ve already given it and asking it to find the latest referral in your Outlook mailbox. In the first case, you might carry on quite happily. In the second, you need that mailbox connection to work too. If the next step is to send a reply or arrange an appointment, those connections matter as well.
That’s the slightly odd position we’re in. Moving some of my work out of email meant I could keep going when the inbox stopped. But connecting an assistant to email, calendars and other systems – precisely so it can do more useful work – can bring those dependencies back into the job. The work has moved, but it hasn’t necessarily become independent. You can buy your AI from a completely different company and still need Microsoft for the particular thing you’re trying to get done. That doesn’t make connecting the systems a bad idea, but it does mean that a different logo on the screen isn’t, by itself, a backup plan.
For me, the useful questions are the practical ones. Can I carry on with the information I already have? Am I waiting for someone else’s input? Will the assistant tell me it couldn’t check the mailbox, or could I mistake an incomplete picture for the whole one?
When the inbox is the front door
I wrote recently about what happens if someone hacks your website. One of the questions bubbling away in that piece was about how much it would matter if the system became unavailable and scaling your risk management to the possible impact. Last week gave us a live demonstration. Some police forces couldn’t receive online crime reports, and some NHS staff went back to paper records.
My eerily quiet inbox was, frankly, a perk, but if you’re running an advice service and email is the only way to make a referral, a quiet inbox can mean people are trying to reach you and you have no idea. Even if the messages all arrive on Monday, the moment to act on some of them may already have passed.
Having other things to get on with doesn’t solve that problem. Neither does having an assistant ready to help if it can’t see the people waiting for a response.
Start with the job, not the tool
I imagine a few people have dug out their contingency plans since then, and they might be wondering whether turning off all the old landlines was such a good idea. If that’s you, and especially if you’re running critical frontline services, I’d start my review with one specific, real piece of work and ask what would happen if one of its connections stopped working.
I wouldn’t start with an inventory of which applications were showing a green light last week – it’ll probably be a different thing that goes down next time. Focus on: could someone actually receive the referral, find the guidance, make the decision and get back to the person, at the pace we need all that to happen?
I’d want staff to be able to tell the difference between “there are no new referrals” and “I can’t check for new referrals”. Those might look much the same on a screen on a quiet Friday afternoon, but I’d ask the supplier to show me what staff see in each case. Different scenarios might not lead to the same decision. And if the fallback is “ask someone to do it manually”, I’d want to know who that someone is, what they need access to and whether that person – or anyone else – has been told. Preferably before Friday afternoon.
If you’re drafting a newsletter or working through a strategy paper, being able to carry on without the inbox for a few hours might be rather useful. For an urgent referral, the missing connection might be the whole point of the job.
I’m still quite happy for my own emails to hold off until Monday. Last week showed me how much I could get done without them. The useful thing to work out is which jobs can carry on – and which ones only look as though they can.