Programmers face the harsh reality of AI

The era of programmer arrogance is over

Programmers who laughed at artificial intelligence a year ago are now seeing layoffs in their own companies. The shift is so rapid that the sector is still struggling to process it: AI not only writes code, it is redefining who is expendable and who stays.

From denial to reckoning

The case of YouTuber and computer scientist Víctor Robles is almost therapeutic. Until a year ago he denied that AI could do much in programming; today his stance has completely changed. He is not an isolated case: in the sector, the denial phase has given way to acceptance, with all the discomfort that entails.

The data nobody wants to see

Stanford's Digital Economy Lab, in a study by Brynjolfsson, Chandar and Chen that cross-references millions of ADP payroll records, has put numbers to the phenomenon: since late 2022, employment for workers aged 22-25 in AI-exposed occupations has fallen 13% in relative terms. In programming, the blow is even harder: employment for developers in that age bracket has dropped nearly 20% between its peak in late 2022 and July 2025. Meanwhile, older workers remain stable or even grow.

The dilemma for consultancies

In IT consultancies, the calculation is stark. If ten programmers used to deliver ten projects a year, with AI agents you can either tell the client they'll be done in a month, or fire nine and maintain the same output. The first option is competitive; the second is too. Companies like Vass have already filed layoffs, and antiestéticar is spreading through departments.

Custom or product: the boundary that decides who suffers

One of the least discussed keys is the difference between custom software and product software. In the former, AI that spits out code is seen as a cost saving, but in the long run maintenance becomes a liability nobody wants to pay for. In the latter, each improvement revalues an asset shared among many clients. AI makes writing cheaper, but not maintaining. That mismatch is what is splitting the job market in two.

Who survives?

There is consensus that junior profiles are the first to fall. Those who only know how to ask AI to fix an error without understanding why it fails become expendable. Those who know how to ask, supervise and distinguish a correct answer from a hallucination remain necessary. But each version of the models improves, and the shelf life of that advantage is shrinking.

The future, in light of this debate, is not an immediate apocalypse, but a painful restructuring: fewer entry-level positions, more pressure on seniors and a new demand for knowing how to supervise the machine. The story is not over, but the era of arrogance certainly is.