"AI Is Removing the Middle Class of Software Engineering" Points to a Sellable Service: Judgment-as-a-Service for Teams Whose AI-Generated PR Volume Has Outrun Their Review Capacity.

Everyone arguing about whether AI is "removing the middle class of software engineering" is missing the part that's actually sellable.

A developer named Florian Herrengt wrote a post that hit the Hacker News front page with 645 points and 545 comments — a comment count that ferocious means it touched a nerve most engineers using AI coding tools are already feeling but haven't said out loud.

His core claim: implementation used to be the bottleneck. A senior engineer's judgment mattered because writing the code was slow enough that judgment had time to catch mistakes before they compounded. Now, per the post, "a person can generate 20,000 lines of code in an afternoon, but you still have to understand what those lines actually do" — and increasingly, nobody has time to.

He opens a PR and it's +24506 -3938 lines, "accompanied by some AI-generated description of what they're supposed to do." Multiply that across a team, and review capacity — not typing speed — becomes the real constraint on whether a codebase stays maintainable.

The business idea

Sell judgment as a service, priced against the exact bottleneck this post names: engineering teams can now generate code faster than any human — including the person who technically wrote it — can review or understand it.

That's a fixed-scope offer: an outside senior engineer (or small team) who reviews AI-generated PRs for architectural risk before they merge, specifically hunting for the failure mode the post describes — decisions that compound because nobody caught them at the point they were made. Not a linter, not a CI check. A human who reads the diff and asks whether the author actually understands what it does, because per the post, engineers increasingly don't: "Didn't you build this like... last week?" now gets answered with a chat log instead of an explanation.

Package it as a weekly or per-sprint retainer — a standing outside reviewer who triages the highest-risk PRs (schema changes, new architecture, anything touching data model or auth) before they ship, for teams whose own senior engineers are already buried.

Who buys this

Teams that adopted AI coding tools early and are now generating PR volume their existing review process was never built for — exactly the population this post is speaking to, and judging by the comment count, a population that's already anxious about it. Fast-moving startups with a thin senior bench are the clearest fit: they have the AI-generated code velocity but not the judgment headcount to match it.

Bottom line

The post's own bet is that "AI pushes salaries further apart" — the best engineers become more valuable, not less, because judgment is what's scarce now. That's not just a hiring prediction. It's a pricing signal: if judgment is the scarce resource, sell judgment, hourly or retained, to the teams whose AI-generated PR volume has already outrun their ability to supply it internally.

Source: https://blog.florianherrengt.com/ai-removing-middle-class-software-engineering.html