The Systems Came Back. But Could We Still Operate?

What a week of technology failures revealed about operational dependency, and the capability we assume is still there

By Louize Clark

Last Thursday evening, Microsoft went down. Not completely, and not for everybody, but for long enough, and widely enough, for millions of people to discover how much of their working day depended on it.

Outlook, Teams and Copilot were all affected. Emails couldn't be sent or received, meetings couldn't happen as planned, and organisations found themselves unable to do things that, on any other Thursday, would have needed very little thought.

It didn't stop on Thursday. The disruption ran through Friday, and Microsoft didn't announce that service had been restored to its last affected customers until early on Saturday morning. By then, users in more than seventy countries had reported problems, and the disruption had reached government departments, courts, police forces, healthcare, financial institutions and businesses.

On the Friday, in the United States, a communications failure in air traffic control led to departures being halted across the entire country, only the second time that has happened since 2001.

These were different incidents with different causes. There's no suggestion they were connected, and that isn't the point I want to make.

What interests me is how quickly an ordinary working day changes when something we've come to rely on is no longer there. And, perhaps more importantly, what that tells us about whether we could still operate without it.

When the ordinary becomes difficult

One of the most revealing parts of the week wasn't the outage itself. It was how people responded to it.

Reuters reported that at one financial firm, trades normally confirmed by email had to be confirmed by fax. Others turned to telephones and Bluetooth. Maersk asked customers to call with anything urgent. And HSBC reminded staff not to use unauthorised channels, such as personal email or WhatsApp, to get around the problem.

That last detail is the one I keep coming back to.

We tend to describe an outage in terms of the technology. The email isn't working. The application won't load. But those are descriptions of the system, not of what happens to the organisation.

An email system being down might mean a solicitor can't receive instructions, a supplier can't confirm a delivery, or a manager can't approve something that twenty other people are waiting on. The same technical failure can be a minor irritation for one organisation and a serious operational problem for another.

And the moment people start finding other ways to get the work done, a different set of questions appears. Are they allowed to use those alternatives? Can they reach the information they need? Are decisions being recorded anywhere? Is sensitive information still protected? Does anybody know which activities have to continue first?

An organisation can have perfectly sensible policies for how people should work when everything is functioning. The harder question is what happens to those policies when following them depends on systems that aren't available. People don't stop working because the approved channel has gone. They find another one.

That isn't an IT problem. It's an operational one, and it goes to the heart of what resilience actually means.

Different doors, same corridor

I don't want to suggest that moving to the cloud, automating processes or relying on large technology providers has been a mistake. Much of what organisations can do today would have been unimaginable twenty years ago. Nor is dependency something we can design out. Every business has always relied on things outside its control: electricity, telecoms, transport, suppliers, people.

What has changed is how much of our ability to function now sits behind a relatively small number of interconnected systems.

A business might use one provider for email, another for its CRM, another for accounts and another for bookings. On paper, that looks like a sensible spread. But those applications may share underlying cloud infrastructure, identity services or integrations that nobody in the business has ever had a reason to look at. The dependency isn't always visible from the name on the invoice.

It's a little like having several doors into a building, only to discover during an emergency that they all lead into the same corridor.

The doors are different. The dependency isn't.

Last week gave us a very practical illustration of that. The outage centred on Exchange Online, Microsoft's cloud email service, but features of Teams and Copilot that depend on it were disrupted too. So an organisation that had spent the past year building AI into its daily work (summarising meetings, drafting replies, pulling together what a client had asked for) didn't just lose its email. It lost the layer it had started using to make sense of its email. The AI wasn't a separate door. It was in the same corridor.

There was another detail in the Reuters coverage that deserved more attention than it got. Organisations that host Microsoft services on their own servers, rather than in the cloud, were somewhat spared. I'm not arguing that everyone should bring their email back in-house. But it's a reminder that how a dependency is arranged matters as much as which provider sits at the end of it.

Earlier this year I wrote about operational dependency as the infrastructure nobody meant to build: the way reliance accumulates through hundreds of reasonable decisions until it becomes something nobody formally chose. That was an argument. Last Thursday was a live test.

And the technology involved wasn't new, experimental or particularly sophisticated. It was email. Software many of us have opened every morning for fifteen years, which is precisely why so few of us had stopped to think about what would happen if it didn't open.

Restoring the system isn't the same as continuing the work

Microsoft's preliminary findings point to a problem introduced during an earlier network capacity upgrade, which surfaced after a more recent network change. Servers became unavailable, the remaining ones were overwhelmed, and some that were brought back failed again as soon as traffic reached them. Microsoft was clear that it wasn't related to cyber activity or AI agents.

That matters, because it reminds us that disruption doesn't need anyone to do anything malicious. Systems fail through ordinary technical problems: a change made months ago meeting a change made last week.

We can't reasonably expect any provider to guarantee that nothing will ever go wrong. What we can expect is that organisations understand what happens to them when it does.

And this is where I think the usual conversation is incomplete.

We measure outages in minutes, hours and days. We talk about restoring access, recovering data and getting systems back online. Those are necessary measures, and organisations invest heavily in them.

But how often do we measure what the organisation can actually do while the systems are unavailable?

Take a business with fifty people. Its main systems go down on a Thursday afternoon. It has a continuity plan, an IT support provider and backups. All perfectly sensible.

Last week, the disruption didn't end on Thursday. It ran into Friday and, for some, into Saturday. So the questions that matter here aren't hypothetical. Could the team see which customers were waiting for something? Could anyone process an order without the system? Who had the authority to approve a different way of doing things? Could the business still take payments, pay suppliers and meet its regulatory obligations?

And if it could, for how long? An hour is manageable. A day is difficult. A week could be commercially damaging.

A backup plan tells you how to restore the technology. It doesn't necessarily tell you whether the people, knowledge, authority and capacity needed to carry on the work are still there.

One detail from the week I found genuinely encouraging. Aviva's chief executive told Reuters that the insurer had tested an outage of a similar scale on its most critical systems about two weeks earlier. That is what it looks like to treat the ability to operate as something you practise, rather than something you assume.

Responsible for it, capable of doing it

There's another part of these incidents that I think gets far less attention than it should.

Microsoft restored its services. Flights resumed. People came back on Monday, caught up on what they'd missed and carried on. That's how these events usually end. The incident is investigated, lessons are discussed, and attention moves elsewhere.

But the dependency hasn't changed. The same systems support the same work, and the people who couldn't work on Thursday were entirely reliant on them again by Monday. If anything, that reliance is still growing, and AI is a big part of why.

An employee who used to prepare a report from scratch now has AI draft it. A customer service team relies on automated summaries to understand the history of an enquiry. A manager uses AI to spot patterns and decide what to prioritise. Each of those changes may be entirely reasonable.

Over time, though, the work reorganises itself around the technology. Roles are redesigned. Expectations rise. People move on. Knowledge that was once exercised every day across a team gets exercised less often, because the system is doing more of it. The organisation can become more productive and less capable of working without the technology, both at the same time.

That isn't inevitable. But it is something we should be watching, because it rarely shows up until something stops.

It also creates a contradiction in how we talk about technology. We encourage organisations to automate, reduce manual intervention and use AI to work more efficiently. Then, when the technology becomes unavailable, or produces something nobody can trust, we expect a person to step in.

But which person?

The employee whose role has changed because the system now does much of what they used to do? The manager who has spent two years reviewing automated reports rather than writing them? The new recruit who has never worked without the software at all?

There's a difference between being responsible for a process and being capable of carrying it out.

I made a similar point recently about human oversight: keeping a person at the end of a process isn't the same as preserving their judgement throughout it. Resilience works the same way. A business can name someone to take over when the technology fails. That doesn't establish whether they can.

People need access to information, opportunities to practise, a working understanding of the task and enough authority to make decisions when normal procedures aren't available. Without those, we risk building contingency plans around capabilities that exist on paper but have quietly disappeared from everyday practice.

And because that disappearance happens slowly, nobody necessarily notices.

Beyond the organisation

There's a wider implication here, particularly for Britain.

I've written before about sovereign AI, and I support the ambition behind much of that work. A country needs secure infrastructure and the ability to make decisions about its own technological future. But I've also argued that building sovereign infrastructure doesn't automatically give organisations the freedom to use it, if their workflows, skills and habits have grown up around somebody else's ecosystem.

Last week adds another layer to that. National resilience can't be measured only by the infrastructure a country owns or how quickly services come back. It also depends on what the organisations operating within it can still do while those services are missing.

The Government said the outage had mainly affected back-office functions, and frontline services such as 999 kept running. That matters, and we should recognise what held rather than treating every disruption as a catastrophe. But think about how many councils, GP practices, schools, care providers, charities and small businesses rely on the same broad technology ecosystem. A care provider that can't access its records, a council that can't reach its residents, a manufacturer that can't process orders: none of those is a national emergency on its own.

Resilience, though, isn't only about individual incidents. It's about how dependency is concentrated across an economy. If thousands of organisations rely on the same underlying services, and many of them have gradually lost the knowledge or capacity to carry on without them, restoring the infrastructure is only part of the answer.

And there's a question of visibility here too. When a major technology provider has an outage, we can usually find out which of its services are affected. What's much harder to establish is what those services support further down the chain. Which organisations depend on them? Which activities can no longer take place? Which of the alternative arrangements rely on the same infrastructure? We may know exactly where the technical problem started without having any clear picture of how far its operational consequences reach. That seems to me a significant gap in how we understand resilience, particularly when so many essential services depend on technology that sits outside the organisations delivering them.

National resilience doesn't exist separately from the organisations, people and services that make up the country. It's built through them.

What we should be asking now

A continuity plan written two years ago may describe an organisation that no longer exists in quite the same form. New systems have arrived. AI features have appeared inside software that was already in use. People have left. Processes have changed, and customer expectations have moved with them. The organisation may have the same name, the same offices and many of the same people, but its ability to operate without particular technologies may be very different.

That's why I don't think operational dependency can be assessed once and filed away. It's part of an organisation's changing operating conditions, and it needs to be revisited as those conditions move.

The systems came back. For many organisations the disruption was temporary, and some had arrangements that worked well. But what if the next interruption lasts longer? What if it affects several services at once? What if the alternative depends on the same underlying infrastructure? Or what if the people expected to take over have spent years working in ways that no longer ask them to use the skills they'd need?

We tend to notice dependency when something stops working. What concerns me more is how much of it develops while everything is working perfectly well.

So before we move on from last week, I'd encourage every organisation to ask itself two questions. If the technology you depend on stopped tomorrow, would you know what else would stop with it? And would your people still have the knowledge, authority and capability to keep the important work going?

Restoring the technology is one measure of resilience. Retaining the ability to operate is another.

Louize Clark is the founder of AI Policies UK, exploring operational dependency, human judgement and the changing relationship between people, organisations and AI. To talk about operational dependency in your organisation, get in touch: Info@aipolicies.uk

Next
Next

When AI Saves Time, Who Gets the Time Back?