In Florian Herrengt’s scenario, a team ships fast with AI coding tools. Then a bug shows up that nobody, not even the AI, can fix. Someone has to go find the one person who remembers how the feature actually works.
That scene appears in a post by software engineer Florian Herrengt, “AI is removing the middle class of software engineering.” He describes a team stuck on a bug for the fourth time, still asking AI to fix it, then finally asking a teammate where the underlying data even comes from. Herrengt writes:
“This project has become so convoluted, with so many layers and services, that no one on your team could possibly start to understand what’s going on.”
His conclusion from that scenario:
“AI makes projects with weak engineering culture fail much faster.”
Who Herrengt is
Herrengt is a software engineer who writes about engineering practice and AI-assisted development on his own blog and in public code repositories.
What he gets right, and where it’s incomplete
Herrengt argues that faster code production magnifies the consequences of engineering judgment. For an engineer who understands the system they’re building, that limit removal is a gift: more of their judgment gets shipped, faster. For an engineer who doesn’t fully understand what they’re building, the same tool multiplies how much unreviewable code piles up before anyone notices a problem.
The review problem is concrete: producing a large change does not mean someone understands it. Herrengt illustrates this with a roughly 25,000-line code change.
Where the post is thinner is on what happens next. Herrengt predicts a wider hiring gap between people who make sound decisions and people who do not. He writes:
“Good engineers have become more valuable because AI lets them move much faster. They don’t need as many people around them just to do the implementation work anymore.”
That’s a prediction about hiring markets and team structure, not something the post demonstrates with data. The post illustrates the argument without measuring a trend across the industry.
What it means for builders
Before you measure AI’s impact on your team by how much code shipped this week, check whether anyone could explain that code to a stranger next month. If the answer is no for a growing share of your codebase, you have a comprehension problem, not a productivity win.
Herrengt’s fix isn’t a tool. He argues for reviewing code, retaining responsibility after a feature ships, and keeping systems understandable. AI makes skipping that discipline faster than ever. It doesn’t make skipping it safe.
End of article