The most repeated prediction of the AI era was that it would come for the coders first. Software engineering, the story went, was the easiest white-collar job to automate, because a model that writes code does not take lunch breaks. Two years into that narrative, somebody finally looked at the hiring data — and the numbers say the opposite happened.
The layer that actually got cut
New hiring data from major tech companies, reported by TechCrunch on June 24, lays out a paradox. Overall tech headcount is down about 25% from 2019 levels. Engineering roles — the supposed first casualties — fell only 11% over the same stretch. And in 2025, a year when every company had every AI coding tool on the market, engineers made up 55% of all new hires at large tech firms.
If AI were quietly deleting programmers, that is not the shape the data would take. The shape it actually takes points one floor up. The role being hollowed out is middle management: the layer that translated executive strategy into engineering roadmaps, ran the status meetings, and turned one person’s plan into another person’s ticket. That layer is down 41%.
Why the robot came for the org chart, not the IDE
There is a logic to this, once you stop assuming “AI writes code” means “AI replaces the person who writes code.” What AI actually compresses is the distance between an idea and a working thing. A single engineer with a decent model can now scope a feature, build it, and read whether it worked — collapsing a loop that used to need a product manager to write the spec, a manager to schedule it, and a second manager to report on it.
The coordination tax is what gets automated, not the building. Middle management existed largely to move information between humans: up to leadership, down to the people doing the work, sideways to the next team. A great deal of that information-shuffling is precisely what software is good at. The human who frames the problem and ships the fix becomes more valuable; the human whose job was to forward that person’s status to someone else becomes a line item.
The catch nobody puts in the press release
This is good news if you write code and bad news if your résumé is a stack of “led a team of.” But it is worth being honest about the asterisks, because this is not a cheerful-tech-optimism publication.
First, “engineering roles down only 11%” is still down 11%, in a sector that spent the entire 2010s only ever hiring. “Resilient” is a comparison, not a guarantee — it means engineers are sinking slower than the people around them, not that the water is draining away.
Second, the cut falls unevenly inside engineering itself. The same dynamic that rewards the engineer who can frame a problem and close the loop is brutal to the engineer who was hired to take tickets and do exactly what the now-deleted manager told them. AI did not make all engineers safer; it made the judgment half of engineering safer and the order-taking half of it disposable. A year ago, those were frequently the same job.
Third, this is the hiring data of the survivors. A 25% drop in overall headcount represents a great many people who never show up in the 55%-of-new-hires figure, because they are not being hired at all.
What to actually do with this
The usable takeaway is not “learn to code and you’re safe.” It is narrower than that, and more demanding. The defensible position in 2026 is being the person who can take an ambiguous problem, decide what to build, build it, and judge whether it worked — without a layer of managers in between converting your output into slides. AI ate the converting. So far it has left the judging alone.
If your value to an employer was that you sat between two other people and passed information along, the data published on June 24 is a warning shot. If your value is that you make the thing, it is — for now — the rare piece of AI-jobs news that points the other way.
