MAJOR OUTAGES · 1 WEEK
7+
ATTRIBUTED TO CYBERATTACK
0
ATTRIBUTED TO AI
0
PERROW · NORMAL ACCIDENTS
1984
In mid-November 2026, a cluster of high-profile system failures hit within days of each other: Microsoft 365, U.S. air traffic control, United Airlines, the NSA's internal network, several major banks, and Apple's services. Authorities said the failures were unrelated—no cyberattack, no single cause. But a former Stripe engineer writing under the name VioletPixel argues the failures share a structural root: the accelerating loss of experienced, deeply knowledgeable people who understand how fragile systems actually work.
Extraction Map
The argument is built on a simple observation anyone who has maintained complex software recognizes. Systems like ACH—the backbone of American banking—are old, flawed, and riddled with legacy debt. They function not because they are well-designed but because specific humans with deep institutional knowledge continuously patch, route around, and manually handle edge cases. Remove those people and the system doesn't degrade gracefully—it fails in ways that are hard to predict and harder to fix. Generative AI enters this picture as the catalyst for removal. To management, the output looks similar: code that compiles, text that reads, interfaces that render. The difference—that the human version came bundled with judgment, pattern recognition, ownership, and the willingness to work an edge case at 2 AM—is invisible from the C-suite. The economic incentive to replace expensive, opinionated engineers with cheaper, compliant AI is enormous and, in 2026, is being acted on at scale across every major technology company. The hollowing-out mechanism is worth naming precisely. It is not just that experienced engineers are laid off. The ones who remain are reassigned from building and maintaining systems to supervising AI output—a role that systematically degrades both their knowledge and their motivation. Every prompt moves them further from the system's internals. Every AI-generated commit they rubber-stamp is a piece of the system they no longer truly understand. The expertise erodes even before the person leaves. The author frames this as capitalism's hot-stove moment: the warning was given, and leadership chose to touch the stove anyway. The framing is blunt but the underlying dynamics are well-documented in reliability engineering literature. Charles Perrow's "Normal Accidents" described exactly this failure mode in 1984—complex, tightly-coupled systems where small errors cascade because the humans who understood the coupling are no longer in the loop. What's new is the speed and scale at which the coupling-aware humans are being removed. The counterargument is real: many systems are still running fine, AI-assisted development can genuinely increase throughput for certain tasks, and not every displaced engineer was irreplaceable. But the author's core point stands on structural grounds. The question is not whether any individual failure was caused by AI displacement—it's whether the aggregate resilience of critical infrastructure is declining as institutional knowledge walks out the door. The failures of November 2026 are data points, not proof. But they are accumulating. What makes this analysis uncomfortable is that the feedback loop is slow. Systems maintained by deep expertise fail rarely and unpredictably. When they do fail, the connection to staffing decisions made months or years earlier is nearly impossible to prove in any single case. This is exactly the kind of systemic risk that markets and management are structurally bad at pricing—until the cascade arrives.
Worth reading?
SKIP
Score
28
/100
Extractive. Value is being captured by firms reducing headcount costs while externalizing reliability risk onto the public, users, and remaining engineers who inherit systems they no longer fully understand.
What Is GRIN?
GRIN stands for Generativity, Resilience, Innovation efficiency, and Novelty openness. These four dimensions capture the essential characteristics of any memetic intelligence — whether a nation, a corporation, a religion, or a movement.
About GRINGenerativity
1/5
Resilience
1/5
Efficiency
2/5
Novelty
2/5
Extractive
4/5
Inverted
Does this create new capability or just redistribute existing value?
Net destruction. Institutional knowledge—the most generative asset in complex systems—is being liquidated for short-term labor savings. No new capability replaces the judgment being lost.
Verified Facts (5)
Author worked at Stripe and learned about banking and money movement systems including ACH
Recent system failures cited: Microsoft 365, Air Traffic Control, United Airlines' systems, the NSA's network, many major banks, Apple's systems
Authorities stated failures are not related, not the result of a cyberattack, and not the fault of generative AI
Article published Tuesday, November 17th, 2026
Author identifies two key traits: intimate knowledge of the system and the drive and determination to force the system to work
More latest news
Read all