AI writes the change. The middle of software engineering dissolves.

The software profession is splitting into a thin senior stratum that designs what gets built and a long junior band that executes what the model produces, and the rung between them is dissolving.

Herrengt's blog post begins with a single number: a 24,506-line pull request, the kind of change a junior engineer could not have written alone and a senior engineer would not have approved in the old workflow. The number is not the story. The story is that the review process that used to make the middle band useful has been deleted. Herrengt's frame is that AI removed the speed limit, which means weak engineering culture no longer fails slowly. The team that used to learn by reading someone else's code now learns by watching a model produce it. The middle of the profession was where comprehension accumulated. AI writes the change. Nobody reads it. Nobody can defend it.

A reader should carry one mechanism forward. When the same bug gets "fixed" four times by prompting an AI, and the engineer who opened the ticket cannot say where the data comes from, the codebase has become something the team can no longer maintain. The bill arrives when customers do. The middle of software engineering was never about output. It was about being the people who understood the system well enough to change it safely. AI is faster at the output. It does not teach the understanding.