If AI 10x's Developers, Why Hasn't the Job Market Collapsed?

How many times have you had this conversation?

Your [insert non technical role here] reads all the headlines from AI labs about 10x and 100x productivity gains and the end of developers. Headlines from people who need to keep the hype train going to fuel their bubble… And then we get the killer line ‘I know this task should take a week but with AI, I am sure you can finish it today’.

Maybe your [insert non technical role here] finally takes the plunge and tries out Claude Code (or equivalent).

They vibe code something over the weekend that vaguely looks like it works. And then we get the killer line. ‘Look what I built in four hours. If I can do this, imagine what a proper developer can do’

I am sick of having this conversation.

So here is the proof that it does not 10x developers.

If it was easy, everyone would already be doing it

I have lived by one saying my whole life: if it was easy, everyone would already be doing it.

Cursor, Copilot and Claude Code are on nearly every developer’s machine already. That part is done. So if a 10x multiplier existed, every company would have it by now. They would need a tenth of the engineers. The job market would be rubble.

So where is it?

Don’t let the headlines do your thinking

We have all seen the tech layoff headlines. Here is the thing: when a big tech firm cuts 5,000 people it makes the front page news. But when the same firm quietly hires 6,000 over the next 18 months, not a peep. Bad news is etched into our memories while good news rarely gets published. That skew is falsely driving a lot of the “AI is replacing developers” narrative.

The US had about 1.69 million employed software developers in May 2025, a record (BLS OEWS), and the Bureau of Labor Statistics projects another 15% growth by 2034 (BLS). The UK government’s 2025 occupations-in-demand analysis counts about 533,000 programmers and software professionals, and flags the role as in elevated demand (GOV.UK).

I admit the data does indicate job postings have fallen. But look at when they fell: 2022, straight off one of the biggest over-hiring bubbles in tech history, in a correction that hit every category of hiring at once. Since early 2025, software postings have climbed about 15% as the wider market shrank 7% (Indeed Hiring Lab). Indeed’s own analysts describe the relationship between AI exposure and postings as “flipping, from job destruction to job creation.”

A 10x productivity multiplier and record developer employment are contradictory.

Fine, then the gain must be hiding somewhere else

Maybe the developers got 10x faster and the workload just expanded to swallow it? Amazon ships faster, so Google has to ship faster to stay relevant, and the 10x becomes the new normal. But then where is the explosion of delivery outputs?

A study of over 100,000 GitHub developers found that autonomous coding agents increased coding activity by 180%. But it decays to 30% by the time you reach actual releases (NBER).

DX measured 400+ companies through the AI boom and found a median pull-request throughput gain of 7.76%. Their own headline: gains are 10%, not 10x (DX).

Maybe every developer on earth is secretly coasting on a tenth of the hours? Then one company defects, works the full week at 10x, and eats its entire industry. Nobody has. And “it’s too early to show up in the data” stopped being an argument about two years ago. The tools have been mainstream since 2023.

Every door a 10x could hide behind is empty.

AI can 10x some tasks, and that is the trap

AI can spit out thousands of lines of code an hour and troubleshoot its own issues. Give it small, isolated, well defined components, and there is a 10x productivity increase right there.

But that definition work is half the work anyway and then you need to spend extra time reining in the mistakes, the hallucinated functions, the confident nonsense. In one randomised trial, experienced developers using AI on their own repositories came out 19% slower, and reckoned afterwards that the tools had made them 20% faster (METR). Even professionals cannot feel the difference between assistance and drag.

It can even 100x some tasks, but it is still a trap

Remember that four-hour demo that [insert non technical role here] vibe coded? In the olden days, this would have taken weeks. So realistically, for non developers it can be a 100x tool allowing them to achieve things they never could previously.

But it only creates thin prototypes which look like they work and writes garbage code which will need to be thrown out. Rapid prototyping and creating reliable good quality software are two completely different tasks.

Be realistic

Why the gains land at 10-30% instead of 10x is a genuinely hard question. It differs by person, team, codebase and organisation, the research is young, and I am not going to pretend to settle it in a blog post.

I don’t need to. The high-level numbers settle the practical question on their own. Employment at record highs. Releases up 30%, not 1,000%. Median throughput gains under 8%. Use the tools, expect a solid boost, and plan your budgets and your headcount around reality.

Save this article so you never need to explain this again. And subscribe at stavra.ai to get the next one.