In 1980, one out of every eight American workers had a job that consisted of producing, typing, filing, and moving written information. Entire floors of office buildings were typing pools. It was a respectable career with a skill at its center. A skill that took years to master, that companies paid real money for, that parents told their daughters to learn.
Nobody held a mass layoff for the typists.
There was no dramatic day where security walked ten thousand of them out of the building. The word processor arrived, then the PC, and companies simply... stopped refilling the seats. Office and administrative support work fell from 12.7% of all U.S. jobs in 1980 to 6.8% by 2022. By 2020, the entire "word processors and typists" occupation was down to 45,200 people, with another -36% projected for the decade. A rounding error, headed to zero.
The last generation of typists thought their skill was the job. It wasn't. It was a feature of the job — and the moment a machine could do the feature, the job quietly reorganized itself around the people who could do everything else.
I keep thinking about them, because I write software for a living, and it's happening again. Same movie. Better graphics.
Nobody is coming to fire you
The scary version of the AI story is a layoff meeting with your name on the list. That version is mostly wrong, and it's wrong in a way that should worry you more.
Here's the version backed by data. Salesforce hired zero new engineers in fiscal year 2026 — Marc Benioff said it on CNBC while the same AI wave shrank his support org from 9,000 to about 5,000. Not fired: not replaced.
Software development job postings are down 36.4% from their pre-pandemic 2020 baseline. A five-year low, still falling 6.7% year-over-year as of late 2025. New-grad hiring at the fifteen biggest tech companies has collapsed by more than half since 2019.
Read those numbers again and notice what they describe. Not an extinction event. An attrition event. A developer resigns, and the req quietly doesn't get reposted. A team of eight becomes a team of five over two years, and the five ship more than the eight ever did. No memo announces this. You only see it if you look at the empty desks, or the postings data.
That's the actual mechanism, and it's the reason the title of this article isn't clickbait. It's the boring, bureaucratic, spreadsheet-driven way that professions actually shrink. Ask the typists.
But the data says something else too, and almost nobody quotes this part
Everyone shares the scary half of the statistics. The other half is more interesting, because the decline isn't hitting "developers." It's hitting a specific enough kind of developer.
Look at the split:
Junior and standard job titles fell 34% while senior titles fell 19%. Employment for software developers aged 22–25 dropped roughly 20% from its late-2022 peak, while workers over 30 in the most AI-exposed job categories grew 6–12% over the same period (Stanford, using ADP payroll data covering millions of workers).
And my favorite statistic in this entire mess, because it's the whole story in two numbers: the U.S. Bureau of Labor Statistics projects "computer programmers" to decline 6% from 2024 to 2034, while projecting "software developers" to grow 15% over the exact same decade.
Same industry. Same wave. One occupation shrinking, its sibling booming. What's the difference between a "programmer" and a "developer" in the BLS taxonomy? Roughly this: one is paid to turn specifications into code. The other is paid to decide what should exist, design it, and be accountable for it running.
The machine came for the first job description. It's actively inflating the value of the second : developers with AI skills now command salary premiums estimated around 43%.
So no, AI is not ending software careers. It's doing something more brutal and more precise: it's unbundling them.
The unbundling
For seventy years, "developer" was a bundle: the judgment (what should we build? is this correct? will it hold?) welded to the production (typing the code). You couldn't buy one without the other, so the market priced them together, and a lot of people built entire careers on being paid mostly for the production half.
More than 84% of developers now use or plan to use AI tools. The production half of the bundle now costs approximately nothing.
And here is the sentence I'd tattoo on the inside of every developer's eyelids:
AI didn't lower the bar for developers. It removed the floor.
There used to be a floor, a level of pure production work below which you couldn't fall, because even routine code took human hours and human hours had a price. That floor carried entire careers. Closing well-specified tickets was a living. Boilerplate was a living. "It compiles and passes QA" was a living.
Mediocrity used to be a living. Now it's a commodity.
But the same unbundling that vaporized the floor put a spotlight on everything the machine can't sell back to you:
Specification — turning a vague human wish into something precise enough to build. This was always the hard part; we just used to hide it inside the typing.
Verification — reading generated code and catching the confident lie in it. I do this every day, shipping production fintech code and building products with coding agents on the side, and I'll tell you the truth the hype skips: the machine's errors aren't sloppy. They're plausible. Plausible errors in code that moves money are how companies die. The person who catches them has never been more valuable.
Judgment — knowing what not to build. The model will generate a beautiful solution to the wrong problem, at scale, forever.
Ownership — being the human accountable at 2 a.m. when the abstraction leaks. No model signs up for that. Someone still has to.
None of these are new skills. That's the twist: the market always paid for them. It just used to buy them bundled with typing, and a lot of us were quietly billing for the typing.
So which side of the attrition are you on?
Run the honest test: does your value survive when the production of code costs nothing?
If what you sell is framework recall, ticket throughput, and code that "works", the empty-desk math is already running against you, and it doesn't need to fire you to win. It just needs to wait.
If what you sell is the unbundled half, or you start building it now, the same math runs for you. Every team that shrinks concentrates its budget on fewer, more trusted people. That's what "+6 to 12% for experienced workers in AI-exposed jobs" means in practice: the money didn't leave the profession. It moved.
Moving with it is not mysterious:
Review more code than you write. Verification is the core skill of the decade. Every generated PR you tear apart is a rep. Ask of every block: what would make this fail?
Write specs like they're the product. Before generating anything, define correct: inputs, edge cases, failure behavior. If you can't specify it, you can't verify it, and someone who can will be paid instead of you.
Ship one thing whole. Design, build, deploy, operate, fix it at night. Ownership can't be learned from tickets, tickets are someone else's judgment, pre-chewed.
Study failures, not tutorials. The machine has memorized every tutorial. Post-mortems, outages, rotten codebases. That's where taste comes from, and taste is what it can't fake.
Supervise the machine like a lead, not a rival. The developers thriving right now delegate the typing and keep the responsibility. That's not a compromise. That's a promotion.
The part the typists would tell you
Here's the thing about the typing pools: the people in them didn't all vanish. Some clung to the skill and went down with it. Others noticed early that the skill was never the point. The office still ran on information, decisions, and someone competent taking responsibility — and they became the administrators, the managers, the operators of the new machines. Same building. Different desk. Better pay.
The profession didn't die. The bundle did.
Software is bigger than typing code, the way offices were bigger than typing letters. The demand for people who can be trusted with what the machines produce is the fastest-growing scarcity in tech, the numbers above say so louder than I can.
They won't fire you. But somewhere in a spreadsheet, it's already been decided whether you'd be replaced. And the criteria have nothing to do with how fast you type.
Move desks.
I'm a senior full-stack developer working in fintech in West Africa. I ship production payment code by day and build AI-powered SaaS solo by night, supervising coding agents — so I'm writing this from inside the split, not above it. If you think I'm wrong, the comments are open. I answer everyone, especially disagreements.
Top comments (0)