Florian Herrengt makes a pointed argument that AI tools rewire the economics of engineering teams — but not through job displacement. The real mechanism is velocity: AI removes the speed limit on implementation. Before, a bad architecture decision took a week to write and got caught in review. Now a single afternoon of agent prompting can produce 20,000 lines of subtly wrong code.
The result is a divergence in engineer value:
- Good engineers become dramatically more valuable — AI multiplies their leverage, letting them skip implementation and focus on architecture, review, and judgment
- Weak engineers become unhireable liabilities — they can now produce bad decisions faster than anyone can review or reverse them, and the complexity snowballs
Herrengt’s diagnosis is that the bottleneck in software has always been understanding, not production. When anyone can generate code at the push of a button, the premium shifts entirely to people who can evaluate whether the generated output is correct, necessary, and maintainable. The team’s most valuable member is the one who doesn’t need to ask “Claude, where does this data come from?” — because they already know.
The essay pairs well with Niklas Gruhn’s [[Don’t be a meat proxy]] (the “meat proxy” is someone who just approves what the LLM produces without adding judgment). Herrengt adds an operational angle: teams where no one understands the full system will hit a wall, and AI won’t help them past it — it will just get them there faster.