Everybody's Lost Their Minds: a security engineer's critique of AI-era work

On September 16, 2026, security engineer and systems administrator Jan Schaumann published an essay titled "Everybody's Lost Their Minds" on his blog Signs of Triviality, arguing that the AI boom has steadily degraded both the quality of technical work and the way the industry discusses it.

Schaumann writes that he now spends more than 75 percent of his working time on AI, directly or indirectly, and that this has cost him most of the enjoyment of his job. He notes that people without an engineering background are pitching "industry changing" products assembled in home laboratories, that professional email reads like social-media influencer copy, and that online writing is converging on sameness.

Vulnerability research and the patching bottleneck

A large part of the essay concerns AI-assisted vulnerability research. Schaumann observes that Anthropic and OpenAI have competed over whose model is more capable and more dangerous, while companies join mysteriously named projects or sign open letters to show they have not been left behind. Engineering effort was redirected quickly: dozens of highly paid security engineers moved to building pipelines able to process thousands of findings.

His central objection is that finding bugs was never the bottleneck in information security. Verifying reports, writing patches and publishing releases are all difficult, he writes, but the hardest step remains getting packages actually updated. Spending the same money on basic hygiene — a complete asset inventory, automated and frequent updates, reboots after roughly 30 days of uptime, attack-surface enumeration across cloud providers — would have delivered far more security in his view.

De-skilling and opaque systems

The essay also questions what happens when AI is layered into development. If a model finds a vulnerability, generates a patch and then "reviews" that patch, the amount of code anyone truly understands shrinks, and the celebrated "human in the loop" is often no more than a rubber stamp. Because debugging is an order of magnitude harder than writing code, organizations running components they cannot explain, he argues, will struggle badly when complex distributed systems fail.

Schaumann is equally critical of companies that ask to be regulated while continuing to build the products in question, and of the environmental cost of the data centers the industry depends on. Human attention, he concludes, remains a zero-sum resource, and individual productivity inside "agentic silos" cannot replace the teamwork that large projects demand.