SAXSKIOLD · Insights · Essay

The Curious Incident of the Dog We Didn't Hear Bark

Complexity Is King — but who is watching the watchdog?

Kenneth Saxskiold-Nørup · 14 November 2026 · 3 min


Hero image placeholder — approved detective and silent dog image awaiting upload.

In Arthur Conan Doyle's Silver Blaze, Sherlock Holmes solves a mystery by noticing something that did not happen. The watchdog didn't bark, and that silence turned out to be the most important clue.

I found myself thinking about Holmes during Microsoft's global cloud outage this week, particularly as the details of what went wrong began to emerge.

Over the years, we have built extraordinarily complex digital infrastructures. And to keep them running, we have surrounded them with equally sophisticated monitoring, intelligent routing, automated health checks, self-healing mechanisms and, increasingly, AI-assisted operations. Technology watching technology, with machines continuously assessing, protecting and correcting other machines.

But what happens when those protective mechanisms become entangled in the very problems they are supposed to prevent?

During Microsoft's outage, servers failed their automated health checks and dropped out of service. The remaining servers inherited the workload, became overloaded and struggled to respond. What makes this particularly interesting is that traffic hadn't increased. The available capacity had simply collapsed, turning what should have been a protective mechanism into part of a much larger cascading failure.

Even parts of Microsoft's own security infrastructure were affected. Safe Links, designed to check potentially dangerous links, could no longer reliably perform that function. The watchdog, it seemed, was standing on the very infrastructure it was supposed to protect.

There is another curious parallel to Holmes' story. The dog remained silent because it recognised the visitor. Microsoft's preliminary explanation also points to something already inside the system: a fault introduced during an earlier network capacity upgrade, exposed by a subsequent change. Not an outside attack, but a weakness that had been there all along.

Restoring stability required engineers to pause deployments, redirect traffic, restart servers, roll back a network change and carefully rebuild capacity. In other words, regaining control meant temporarily limiting the complexity and automation on which normal operations depend.

And perhaps that is where the larger lesson lies.

We tend to respond to technological complexity by adding more of it. More monitoring, more automation, more intelligence and more layers of control. Yet every additional layer introduces its own dependencies, and every system designed to protect another system must ultimately depend on something itself.

So how do we know whether the watchdog has stopped barking, or whether we have simply lost the ability to hear it?

Microsoft has ruled out AI agents as the cause, and its final incident report is still pending. I intend to return to the technical details when it becomes available, but the broader question is already worth considering.

Have we made our infrastructures so dependent on automated control that the ability to regain human control has become one of the most important measures of resilience?

Perhaps the future of intelligent infrastructure is not simply about achieving greater autonomy, but about designing what I would call recoverable autonomy: systems that remain understandable, manageable and ultimately controllable, even when their own automated safeguards no longer work as intended.

Because the real test of an intelligent system may not be how independently it operates when everything works, but whether we can still take control when it doesn't.

Sources

The PDF opens in your browser’s print window — choose “Save as PDF”.

More insights

All insights →

From the Field is a short monthly letter from CURRENT. Receive it.