Last week, in a WhatsApp group, a friend sent me the question that won’t leave the head of anyone in IT: “so, will AI replace programmers?”. Everyone expected me, someone who makes a living from this, to calm the crowd down. Well, I’ll do the opposite: I think it will. Programming, the act of writing code, will be entirely replaced. But hold on, drop that imaginary layoff letter, because the story does not end there.
My thesis is simple: the code goes, IT stays. And there is an old scene from our own history that explains this nicely. Hang on, I’ll explain.
What the studies say about AI and programming
First, the facts, because this debate is full of guesswork on both sides.
On the “AI is unbeatable” side: a controlled experiment published by researchers from MIT, Princeton and Microsoft showed that developers using GitHub Copilot completed a task 55.8% faster than the group without the tool. More than half the time, gone.
On the “hold your horses” side: a randomized study by METR, from 2025, took experienced developers working on their own open source projects and found the opposite: with AI they were 19% SLOWER, and the funniest part is that they believed they had been 20% faster (the illusion is real, it happens rs).
And there is the finding that impressed me the most: the study “Canaries in the Coal Mine?”, by Erik Brynjolfsson and his team at Stanford, analyzed payroll data from millions of American workers and found a 19% employment drop for young people aged 22 to 25 in the occupations most exposed to AI (programming included), while experienced professionals are doing just fine. The entry door is closing first. And the study itself makes a distinction that is the heart of this article: employment falls where AI substitutes the task, and holds or grows where it complements the work.
Hang on, I’ll explain: the spreadsheet and the bookkeeper
Do you know which profession the electronic spreadsheet killed? Not the accountant’s. It was the bookkeeper’s, that professional who spent the day entering numbers by hand into huge ledgers. The calculation itself became a commodity. But the accountant, who interprets the numbers, signs off and answers for them, has never had so much work.
Code is the new ledger entry. Writing the function, the CRUD, the unit test: AI already does that, and it will do it better and cheaper every year. Insisting on competing with it there is like competing with the spreadsheet at entering numbers. Seriously, stop and think: do you want to be the bookkeeper or the accountant of this story?
That is why I say programming itself will be entirely replaced. It is not an alarmist hunch; it is looking at where the evidence points (and at my own routine: I have already written here about vibe coding and git worktree (in Portuguese), with agents writing code in parallel while I decide what gets in).
The future of programming: what remains when code writes itself
If the act of typing code goes away, what remains? Everything that comes before and after the code. Truly understanding the problem, choosing the architecture, setting boundaries, reviewing, integrating, answering for the result. A World Economic Forum analysis from January points exactly there: 65% of developers expect their role to be redefined toward architecture and AI-supported decision-making, with the repetitive part automated.
Notice it is the same distinction from the Stanford study: substitution in the task, coexistence in the profession. The developer who only turns tickets into code is on the substitution route. The engineer who turns business problems into solutions, and uses AI as an entire team of tireless bookkeepers, stays.
Don’t compete with AI at the task; command AI in the profession.
IT jobs that coexist with AI
And the other IT areas? This is where it gets more interesting, because several of them are made precisely of what AI does not deliver on its own: context, accountability and trust.
Information security is the most obvious example: the more AI-generated code, the more attack surface and the more auditing needed (who is going to check what the agents wrote?). Architecture and platform engineering stay strong, because someone has to decide where all of this runs and how much it costs. Data and product too: AI answers, but the right question is still human work. And the OECD itself, which estimates 92 million jobs lost to AI worldwide (in Portuguese), shares the stage with the World Economic Forum projecting twice that number in new roles. Destruction and creation, at the same time.
This is not a guarantee for anyone to sleep easy, let’s be clear. It is a map: the areas that coexist with AI are the ones that carry judgment and accountability, not typing. I have already written about the essential skills for 2030 (in Portuguese), and well… that list has aged nicely.
AI sweeps the task; it does not sign for the responsibility.
Frequently asked questions
Does AI already replace programmers today?
At the task level, yes: routine code generation, tests, refactoring. At the profession level, not yet: the METR study showed that even experienced developers get slower with AI on complex code. But the effect already shows up in entry-level hiring, as Stanford demonstrated.
Is it worth learning to code in 2026?
It is, but with a different goal. Learning to code today is like an accountant learning accounting: you need to understand what the machine does in order to review, decide and take responsibility. What changed is that typing code is no longer the final product.
Which IT areas are safest from AI?
The ones AI complements instead of substituting: information security, software architecture, platform engineering, data and product. The common pattern: judgment, business context and accountability for the outcome.
Don’t the studies contradict each other (55% faster vs 19% slower)?
They measure different things: an isolated, simple task in one case, a real, complex project in the other. To me, the two together tell the same story: AI already owns commodity code and still stumbles on context. Guess which half of the work is left for us?
Code goes the way of the bookkeeper; those who think, decide and sign stay.
Cheers, Fellipe Soares