I'm going to make a claim that's going to upset some people, including some people I respect: most of the "AI-driven layoff" narrative in 2026 is bullshit, and we're letting CEOs use it as cover for a different story.
I want to be careful about what I am and am not saying. AI is real. It is changing work. It will continue to change work. None of that is in dispute. What I'm saying is narrower: the causal chain being sold in press releases — "AI made us productive, so we don't need these people" — is mostly not supported by the productivity data we actually have. And I think we owe each other more honesty about that.
Two pieces dropped this week that crystallized it for me.
The data and the story stopped matching
NBC ran a piece summarizing METR-reported findings: experienced developers were about 19% slower on real tasks when using AI tools, even when many of them believed they were faster. Same week, you could refresh LinkedIn and watch a parade of CEOs frame layoffs as AI-efficiency outcomes.
These can't both be straightforwardly true. Either AI is making developers faster — in which case the slowdown evidence needs an explanation — or it isn't, in which case the "we don't need these people because AI" framing is doing something other than describing reality.
The Conversation took the framing apart and arrived at roughly the read I've been carrying: most "AI layoffs" are post-ZIRP headcount correction plus investor signaling, with a thin layer of AI narrative laid on top because that narrative is socially and financially cheaper than admitting "we over-hired in 2021 and the cost of capital changed."
Why the AI narrative is the convenient one
Sit in a CEO's chair for a minute. You over-hired in 2020–2022 when money was free. Now money isn't free. You need to cut 10–20%. You have three explanations you can give the market:
- "We made a strategic error." Stock punished. Board annoyed. Your tenure shortens.
- "Margin pressure from competition is forcing this." Stock punished. Suggests weakness.
- "AI is making us more efficient." Stock rewards. You look forward-looking. You're not cutting — you're transforming.
Option three is not a lie, exactly. But it's not a careful description of the causal chain either. It's a rationalization that happens to also be the most market-friendly explanation. If you wonder why it's the explanation we keep hearing, this is most of the reason.
Look — I don't think every executive saying this is being cynical. I think a lot of them have genuinely convinced themselves the chain is real. AI did enter the workflow. Layoffs did follow. The brain pattern-matches a story. That's how humans do narrative.
But the data is what the data is. And until I see a serious peer-reviewed study showing sustained, broad-based productivity gains in real engineering work — not vibes, not vendor white papers, not "developers said they felt 20% faster" surveys — I'm going to keep my hand on the bullshit detector.
What I think is actually happening
Here's the version I'd defend:
Compression, not replacement. The labor signal isn't extinction. It's compression. Fewer entry and mid roles. Sharper premium on engineers who can actually ship AI in production. A flattening of the career ladder where the rungs that mattered most for early-career growth are quietly being removed. That's painful and serious and worth talking about. It's also a different problem than "AI is doing my job."
Pre-existing trends getting AI-labeled. Customer support reductions have been creeping for a decade, driven by chatbots and self-service before LLMs. The "AI replaced our CS team" framing is half true and half a marketing relabel of a slow trend that finally hit a tipping point. The trend is real. The "this just happened because of AI in 2026" framing is not.
Productivity gains that exist but are uneven. I'll concede AI helps materially in some workflows — boilerplate code, routine documentation, repetitive triage. It hurts in others — complex debugging, novel system design, anything heavy on tacit context. The average is unimpressive. The variance is huge. The story-tellers conflate the helpful slices with the average and sell the average.
The talent that benefits is concentrating. Engineers who already had strong system context, judgment, and integration skill are getting a real multiplier. Engineers earlier in the curve are not. So the productivity story is more about which engineers than about whether engineering work overall is faster. That's a much less marketable framing for a press release.
What I want to be wrong about
Let me steelman the other side honestly, because I might be too cynical about this.
The strongest pro-narrative argument I can think of: maybe the productivity studies measure the wrong things. METR-style task experiments are bounded by design. They may miss the compounding gains — code reuse, faster onboarding, lower bug rates downstream — that show up in quarterly metrics but not task-level ones. A team that ships 19% slower on tested tasks but has 30% fewer regressions in production is not actually slower. It might be a lot faster.
That's plausible. I'd take the argument seriously if I saw the longitudinal data. So far, what I see is short-horizon studies showing mixed-to-negative results, plus executive narrative going strongly positive without the data to back it.
Where I want pushback: if you've run a careful before/after measurement on AI tools at your team or company, what did you find? Especially the boring middle case — the one where it's modest, complicated, and doesn't fit the hype or the doom narrative.
If you're an engineer and you're scared right now
Two practical things I'll say with confidence.
One: stop measuring your job security by the noise. The noise — executive quotes, layoff headlines, doom Substacks — is downstream of business pressures that have very little to do with your actual marginal value. Your value as an engineer is set by your team's outcomes, not by the AI narrative cycle.
Two: bet on the skill premium going up, not down. The compression case I described is bad for engineers in the middle who haven't yet picked up the higher-leverage skills — system thinking, deployment judgment, agent orchestration, the integration work I keep writing about. It is good for engineers who do. That premium is not going away. If anything, it's getting steeper. Aim there.
I am not telling you to ignore the layoffs. They are real and they hurt real people. I am telling you the framing matters. If you internalize the "AI is eating my job" story without examining it, you're going to make worse decisions about where to invest your time. The honest version of the picture — labor is compressing, the top tier is doing fine, the middle is squeezed — leads to better choices than the panic version does.