I've been following the ongoing Microsoft 365 and Exchange Online incident since yesterday.

And I have to admit, I've been thinking quite a bit about the people working behind the scenes.

Not because Microsoft is a competitor in parts of our business, or because there's some commercial advantage in watching another provider experience difficulties.

Quite the opposite.

As someone building and operating cloud infrastructure myself, I know that even a relatively small incident can become complicated very quickly. Multiply that by Microsoft's global infrastructure, the number of interconnected services and the enormous number of people depending on them, and the operational challenge becomes difficult to comprehend.

I genuinely hope the teams involved are doing well.

But the incident also provides an opportunity to discuss something that rarely receives much public attention.

What happens to decision-making, coordination and human performance when a major infrastructure incident continues for hours, under pressure and with no simple solution in sight?

There is actually a great deal of research on this. And some of the findings are both fascinating and surprisingly relevant to the people keeping our digital world running.

First, what do we actually know?

Let's separate the facts from the assumptions.

Microsoft's published service-health communications confirm that incident EX1550543 began on November 12, 2026, affecting access to Exchange Online mailboxes through several connection methods, including Outlook on the web, MAPI, Exchange ActiveSync and REST APIs.

A related incident, MO1550821, documents disruption to other Microsoft 365 services.

Microsoft reported that a significant number of servers had unexpectedly become unavailable, shifting traffic onto the remaining infrastructure.

During recovery, some servers initially returned to service but became unresponsive again when they began receiving traffic.

The company subsequently described a process responsible for establishing secure connections becoming overwhelmed.

Its reported mitigation efforts have included redistributing traffic, temporarily limiting certain connections, isolating affected infrastructure, restarting components and testing configuration changes.

As of the morning of November 13, Microsoft reported increased stability after applying a patch to an upstream authentication and routing configuration component. The company was systematically expanding that change while continuing to monitor recovery.

The incident remained open in the latest verified update.

Microsoft has stated that it is not aware of malicious activity causing the incident. A final root-cause analysis has not yet been published.

These details are important because they illustrate the complexity of the recovery process. But they do not tell us how Microsoft's engineers are feeling, what internal disagreements may have occurred, or how its crisis organization is operating.

We should not pretend otherwise.

What we can explore is how major incidents are generally managed, and what research tells us about the human factors involved.

The first challenge: Fighting the incident without making it worse

When an infrastructure service fails, the instinctive question is often:

What caused this?

But in a serious incident, there may be a more urgent question:

What can we safely do right now to reduce the impact?

Finding the underlying root cause and restoring service are related, but they are not always the same task.

Consider a hypothetical situation where a group of servers becomes unresponsive.

You restart them.

They recover.

Traffic returns.

And then they fail again.

Suddenly, the obvious solution is no longer so obvious.

Restarting more servers may increase available capacity, but it could also recreate the failure. Moving traffic away from one group of servers may protect those machines while increasing the load elsewhere.

Even a technically correct recovery action can have unintended consequences in a complex distributed environment.

This is where established incident-management procedures become essential.

Experienced teams generally aim to contain the impact, protect data integrity, make changes in a controlled manner and verify whether each intervention produces the expected result.

Sometimes restoring service gradually is safer than restoring everything at once.

A slow recovery is not necessarily an indication of poor incident management. It may be the result of deliberately avoiding a second failure.

And that requires something that can be difficult when thousands of customers are waiting: patience.

Tunnel vision: What happens to our thinking under pressure?

One of the most interesting aspects of crisis psychology is how stress can affect cognitive performance.

In 2016, Shields and colleagues published a meta-analysis examining the effects of acute stress on core executive functions.

They found that acute stress impaired working memory and cognitive flexibility, while its effects on inhibitory control were more nuanced.

Working memory allows us to hold and manipulate information while solving a problem.

Cognitive flexibility allows us to switch between perspectives, reconsider assumptions and adapt when circumstances change.

Both are enormously important during complex technical incidents.

Imagine an engineer investigating an unexpected failure.

The available evidence initially suggests that a recent deployment is responsible.

That becomes the working hypothesis.

New information arrives. Some of it supports the hypothesis, but other observations suggest a different explanation.

Under pressure, it can become more difficult to keep competing explanations in mind and continuously reassess the situation.

This is where tunnel vision becomes relevant.

A 2008 study published in the journal Human Factors, titled Improving Decision Making in Crisis Response through Critical Thinking Support, examined problems including tunnel vision, information bias and inadequate communication of reasoning between crisis-response teams.

The researchers explored methods for making hypotheses and the reasoning behind decisions more explicit.

An important lesson is that teams need to communicate not just what they believe, but why they believe it.

There is also an important distinction here.

Stress does not automatically make skilled professionals ineffective.

Experience, training and familiarity with emergency situations can support rapid, high-quality decisions. Research into naturalistic decision-making shows how experienced responders recognize patterns and apply learned strategies under demanding conditions.

The challenge is knowing when a familiar pattern no longer adequately explains what is happening.

One useful question in any incident is:

"What evidence would prove our current hypothesis wrong?"

It sounds simple.

But asking that question at the right moment can be extremely valuable.

This is a general principle of crisis decision-making. It is not an assertion that Microsoft's teams have experienced tunnel vision during the current incident.

The danger of needing to do something

There is another human tendency worth discussing.

When something serious happens, doing nothing can feel unacceptable.

Customers are waiting. Management wants updates. Monitoring dashboards continue showing errors.

Every minute feels expensive.

That creates pressure to act.

But activity and progress are not the same thing.

In incident management, an intervention should ideally have a clear purpose, a measurable expected outcome and, where feasible, a recovery or rollback strategy.

If several major changes are made simultaneously, determining which one helped or worsened the situation becomes much more difficult.

This is why a structured technical response is so important.

Instead of allowing every available engineer to independently modify production infrastructure, mature incident-management models coordinate operational changes through defined roles.

Google's Site Reliability Engineering documentation describes a model derived from the Incident Command System, separating incident command, operational work, communications and planning.

The Incident Commander maintains overall situational awareness and coordinates priorities.

The technical responders concentrate on diagnosis and mitigation.

Communications personnel keep stakeholders informed.

Planning supports documentation, resources and handovers.

The structure can be adapted to the size of the incident and the organization.

But the underlying idea is powerful.

The people repairing the infrastructure should not also be responsible for managing every consequence of the outage.

Fatigue: The problem that becomes more important over time

Now consider what happens when an incident lasts through the night.

Even when engineers work in shifts, a prolonged incident creates challenges involving continuity, concentration and information transfer.

Researchers have repeatedly studied how insufficient sleep affects decision-making and error detection.

A 2021 systematic review and meta-analysis in Sleep Medicine Reviews examined the effects of sleep loss on performance monitoring and error monitoring.

The findings supported impairment in error monitoring, although the wider evidence on people's awareness of their own performance was more mixed.

This matters because making a mistake and detecting that mistake are two different abilities.

An engineer may still be capable of performing familiar technical tasks while becoming less effective at noticing unexpected outcomes.

A 2025 review of sleep loss and decision-making also highlights how the effects can differ across cognitive processes and may not always be apparent from simple performance measurements.

None of this establishes that Microsoft employees are sleep-deprived. We have no evidence about their shifts, rest periods or staffing arrangements.

But it demonstrates why effective incident management must account for fatigue rather than relying indefinitely on individual endurance.

A proper handover is not merely telling the next engineer what is broken.

It should communicate what is known, which assumptions remain unverified, which changes have been made, what the current risks are and which actions must not be repeated without further evaluation.

The incoming team needs enough context to continue safely, without having to reconstruct the entire incident.

And perhaps most importantly, engineers must be able to step away without feeling that they have abandoned their colleagues.

A resilient organization should not require exhausted people to become heroes.

Psychological safety: Can someone challenge the expert?

Another fascinating area of research concerns psychological safety.

Amy Edmondson's influential 1999 study, Psychological Safety and Learning Behavior in Work Teams, examined how teams learn and communicate.

Her research found that psychological safety was associated with learning behavior.

In practical terms, psychological safety concerns whether people believe they can raise concerns, admit uncertainty or challenge assumptions without being punished or humiliated.

Think about what that means in an incident room.

A senior engineer has spent several hours investigating a problem and is confident about the likely explanation.

A less experienced colleague notices something that does not fit.

Will that person speak up?

And if they do, will the observation be taken seriously?

A team can contain extraordinarily talented individuals and still make poor collective decisions if important information is not shared.

This is particularly relevant during an emergency, when time is limited and confidence can easily be mistaken for certainty.

The objective is not to eliminate leadership or debate every decision indefinitely.

It is to create an environment where relevant information can challenge existing assumptions, regardless of who provides it.

An Incident Commander should have the authority to coordinate the response without becoming immune to criticism.

A senior engineer should be able to say, "I was wrong."

And a junior engineer should be able to say, "I think we're missing something."

That is not a weakness in the command structure.

It is a form of organizational resilience.

The invisible responsibility of incident leadership

I think this is one of the aspects of incident management that outsiders rarely appreciate.

The Incident Commander is not necessarily the person who knows the most about the affected technology.

In many situations, the most valuable technical expert should be allowed to remain focused on technical work.

The command role involves maintaining situational awareness, making sure responsibilities are clear, coordinating resources and protecting the team's decision-making process.

It also means managing interruptions.

Imagine several executives repeatedly asking the same engineers for an estimated recovery time.

Each question may be understandable.

But collectively, those interruptions consume attention that should be directed toward resolving the incident.

Good communication procedures allow leadership, customers and other stakeholders to receive meaningful updates without constantly interrupting the technical response.

There is also a psychological dimension to this.

An effective leader should be able to recognize when the team is becoming overloaded, when assumptions need challenging, when additional expertise is required and when individuals need to be relieved.

This is one reason incident response cannot be reduced to technical expertise alone.

It is a combination of engineering, coordination, judgment and human factors.

When is the incident actually over?

Another question that sounds straightforward but often isn't:

When can we safely declare the service restored?

In a distributed system, availability can return gradually.

A server may pass its health checks but fail when exposed to ordinary production traffic.

A user may successfully open an email application while another remains unable to connect.

A service may be technically online while message queues, dependent applications or background processes are still recovering.

Recovery therefore requires evidence.

Are real user transactions succeeding?

Is the infrastructure stable under expected load?

Are backlogs decreasing?

Are errors returning to normal levels?

Are dependent services recovering?

And perhaps most importantly: does the system remain stable after traffic is reintroduced?

These questions help distinguish a temporary improvement from sustained recovery.

Microsoft's reported experience with servers becoming unresponsive again after receiving traffic demonstrates why this distinction matters.

The objective is not simply to make a dashboard turn green.

It is to restore reliable service.

What happens after everyone goes home?

Once an incident is resolved, there is still important work ahead.

A good post-incident review should examine not only the technical failure but also the conditions that shaped the response.

Which safeguards worked?

Which dependencies proved more important than expected?

Did monitoring provide useful information?

Were changes coordinated effectively?

Did responders have enough context?

Were staffing and handover arrangements sufficient?

What can be improved before the next incident?

I strongly believe in blameless postmortems.

Not because accountability should disappear, but because an investigation focused primarily on finding someone to blame can discourage the openness needed to understand complex failures.

Accountability and a learning culture are not mutually exclusive.

People should be responsible for following established procedures and making defensible decisions.

Organizations, meanwhile, must take responsibility for the systems, incentives, operational constraints and working conditions that influence those decisions.

A postmortem that simply concludes someone should have been more careful is rarely enough.

The useful question is how the system can become safer even when humans inevitably make mistakes.

A reflection from a much smaller cloud provider

At WAYSCloud, we're building an independent cloud platform, including meil.no, our email and collaboration service.

Our scale is nowhere near Microsoft's.

Not even remotely.

And I'm not going to pretend that operating a smaller platform gives us some special immunity to complex failures.

It doesn't.

But some of the underlying responsibilities are the same.

Customers depend on our services. They trust us with their data. They expect us to make responsible decisions, including when things go wrong.

That means thinking about resilience before incidents occur.

It means designing recovery procedures, understanding dependencies, testing assumptions and creating a culture where engineers can challenge one another constructively.

It also means remembering that reliability is not just an engineering metric.

There are people behind it.

People who need rest, clear responsibilities, good communication and the confidence that their organization will support them when difficult decisions must be made.

And there will be incidents. Every serious infrastructure provider has to prepare for that reality.

The objective is not to promise perfection.

It is to build systems and organizations capable of responding responsibly when perfection proves impossible.

To the people working behind the scenes

There will be time to evaluate the Microsoft incident in detail once the facts are available.

There should be.

Services of this importance deserve transparency, rigorous technical scrutiny and meaningful lessons learned.

But criticism of systems and respect for the people operating them can coexist.

I think it's worth remembering that the individuals working to restore these services are fellow engineers, incident responders and technical professionals.

They are not the outage.

And we should not confuse the existence of a serious incident with a lack of dedication or competence among the people responding to it.

For those of us building alternative platforms, there is no contradiction in competing with Microsoft while sincerely wishing its teams well.

In fact, I think that kind of professional respect is important.

The real measure of an engineering culture is not whether its people can work endlessly under pressure. It's whether the organization gives them the structure, support and freedom to make good decisions when things are at their most difficult.

To everyone involved in restoring Microsoft's services: I hope the recovery goes safely, the lessons are useful, and that you all get some well-deserved rest afterward.

From one cloud builder to another, you have my respect.


Sources and further reading

Microsoft incident information

Incident response and human factors

  • Google, Site Reliability Engineering: Managing Incidents. Read
  • Google, Site Reliability Engineering Workbook: Incident Response. Read
  • Decision-Making During High-Risk Events: A Systematic Literature Review (2023). Review of 32 empirical studies. Read
  • Shields et al. (2016), The effects of acute stress on core executive functions: A meta-analysis and comparison with cortisol. PubMed
  • Improving Decision Making in Crisis Response through Critical Thinking Support (2008), Human Factors. Read
  • The impact of sleep loss on performance monitoring and error-monitoring: A systematic review and meta-analysis (2021), Sleep Medicine Reviews. Read
  • The impact of sleep loss on decision making: Opening the cognitive black box (2025). PubMed
  • Edmondson, A. (1999), Psychological Safety and Learning Behavior in Work Teams. Administrative Science Quarterly, 44(2), 350–383. Read

Note: This article reflects publicly available information verified on November 13, 2026, including Microsoft's service-health update at 08:06 UTC. The psychological and operational research describes general principles, not assessments of individual Microsoft employees or undisclosed internal incident-response procedures.