Why AI hasn't replaced software engineers, and won't

That basic view landed with a lot of people, but in a narrower form than the post presents. The strongest consensus was that AI already crushes boilerplate, glue code, greenfield prototypes, and many internal tools. People running agencies and in-house teams said work that took five people now often takes one or two, especially for early-stage projects. The bottleneck has shifted up the stack. The job is increasingly translating messy business needs into constraints, steering the model, reading what it produced, and catching the failures the model cannot see. Several commenters said this is not a small change. It is a real compression of headcount even if “software engineer” as a category survives. Where the conversation got sharper was maintenance. Again and again, people pointed out that the hard part was never getting a demo on screen. It is adding the 433rd feature without breaking the first 432, surviving incidents, staying secure, handling regulation, and keeping a messy codebase coherent over time. A few concrete anecdotes made this point stick. Non-technical teams can now ship something in days, but months later someone is buried in fixes, architecture problems, and accidental complexity. In practice, many “AI replaced engineers” stories were really “AI created software that now needs an engineer-shaped adult in the room.” Comments also split on where AI bites first. A lot of people said backend and infrastructure work are more automatable because they are text-heavy, testable, and easier to loop on with tooling. Others pushed back that frontend commodity work is exposed first because the stakes are lower and modern design systems make page assembly increasingly mechanical. The common ground was that the easy, repetitive, and low-accountability parts of the stack are going first. Taste, cross-system judgment, and domain context are holding up better than syntax production. The mood was not calm optimism. It was more like grudging acceptance that AI is genuinely useful, paired with skepticism toward sweeping “nobody loses” stories. Even people bullish on long-term software demand expected pain to hit unevenly. Junior and mediocre developers were seen as most exposed. Strong senior generalists with domain knowledge were seen as more valuable because they can supervise multiple agents and own outcomes. A few commenters went further and argued pay will polarize rather than simply fall, with elite engineers becoming more leveraged while low-end commodity coding gets cheaper or disappears. Others thought that is too rosy and that overall bargaining power still weakens when the same output needs fewer people. A final thread challenged the post’s strongest “won’t” claim. Several commenters said the article quietly assumes current limitations are structural when they may just be temporary capability gaps. They accepted that today’s models still need heavy supervision, but rejected the leap from “not autonomous now” to “won’t replace engineers.” The more grounded position that emerged was simpler: AI is already changing software work dramatically, but today it looks much more like compression, polarization, and expansion of software volume than a clean disappearance of the profession.