Why AI threatens far more jobs than just programmers

AI threatens doctors and lawyers, not just programmers

There's a scene that keeps repeating itself: someone opens YouTube and finds nothing but outraged programmers. AI writes bad code, they say, supervision is lacking, they spent decades learning every language only to watch a model produce code in seconds. The complaint is loud. And yet the work AI turns into trivia —tabulating data, repeating a lesson, reading a medical test, filling out a legal form— occupies millions of people who barely raise their voices.

Behind it all lies an uncomfortable question: if programming was the peak of complexity and it's already being automated, what's left below it? The topic that has dominated the conversation in recent weeks points to the same place: the threat isn't measured by technical difficulty, but by how much of the job is repeatable procedure.

Who shouts and who stays silent about AI

Programmers have something others don't: their own megaphone. For years they sold courses on languages and frameworks, and now they've rebranded them as courses on AI and agentic programming. The catalog shift has a number behind it: on platforms like Udemy, new courses available would have dropped by around 90%, with sales of the old ones heading the same way. When the teaching business dries up, it's worth asking what was being taught and why.

In graphic design and photography the debate is older. Some argue generative AI has put many professionals on the ropes; on the other side weighs the argument that most already use it, consciously or not. The interesting nuance: AI-generated designs, precisely because they're infinite and free, tend to look alike. If you want something unique, you'll have to pay for it.

Translators are a special case. For a court, a document must be signed by a sworn translator. The consequence isn't mass unemployment but concentration: fewer professionals, better paid. It's the pattern repeating itself in almost every affected trade, and it underpins the central thesis: jobs are lost, the profession is not.

Who signs the prescription? The shield of liability

The most repeated argument against total replacement is the signature. In Spain there are around 20,000 procuradores (court procedural representatives) whose real job is moving paperwork between institutions, plus a legion of notaries and registrars who exist, to a large extent, to authorize things. Who bears civil and incivil liability if AI makes a mistake validating something is the question that stalls entire projects.

The opposing answer is just as blunt. Civil liability —where? How many lawyers, prosecutors or judges have personally answered for a mistake. If the problem is insurance, you buy it: a liability policy subscribed by millions of users comes cheap. What today looks like a legal wall becomes, with enough volume, a monthly premium.

How many jobs has AI already destroyed?

It's the question that disarms the alarmists. Since ChatGPT burst onto the scene in 2022, job losses directly attributable to AI are hard to find. What it has created is an army of trainers, consultants and commentators on the phenomenon itself.

The pessimistic scenario isn't about mass layoffs, it's about prices. The model the most skeptical use is simple: if there are 100 professionals and the work shrinks to 20, the 40 who retrain push from one side and the 40 trying to return push from the other, all downward. The result isn't outright unemployment, it's a collapsed salary floor. No one has yet measured how long that adjustment takes or which sectors get hit first.

The cost nobody includes in the calculation

Automation isn't free. Every inference consumes tokens, and even though the unit price keeps falling, consumption explodes with intermediate processes. Some companies have already shelved projects over this: the budget runs out solving trivialities. The image used is that of the arcade machine and its INSERT COIN.

On top of that comes dependency. Putting a critical process in the hands of an external provider means exposure to any price change. And in parallel, prices will eventually be capped by Chinese models, which play in a different league on cost. Running the same work locally, without depending on third parties, is cheaper than sending every document to a cloud service to be reviewed and formatted.

Robots, mathematics and the rest of the chain

There are signs that go beyond software. Chinese humanoid manufacturer AgiBot would have boosted its shipments by 562% and would hold 44% of the global market, driven by industrial demand. In Madrid, driverless metro trains already run and autonomous taxis are making pogre.

And in fields thought to be reserved for human genius, an analysis of open mathematical problems credits AI with solving five of the last six cases, with a sixth solved with human help. There, supervision is no longer the debate: the debate is whether anything is left that the machine can't reason about better than the person who programmed it.

The exact point where all this gets stuck is the timeline. Nobody disputes that AI will reach almost everything; what nobody has managed to calculate is when, and who will have to open the door first.