Anthropic CEO predicts AI will write all code in a year

AI to write all code in 12 months, say CEOs

Are we really one year away from artificial intelligence writing 100% of the code? That is what the top leaders of Anthropic and OpenAI claim, setting a date for the end of the human programmer. "This is the year AI becomes better than humans at programming forever. There is no going back," states OpenAI's CPO. The technical community does not believe it, and they have good reasons.

The promise is not new. Twelve months ago, similar claims were made, yet AI-generated code still requires human oversight. "If you needed 5 programmers before, after this you will need 3 or 4. Is that pogre? Yes. But using AI generates tons of errors, and if you don't know how to program, there is no way to fix them," summarizes an active developer. The gap between the headline and keyboard reality is, for now, an abyss.

The bottleneck is not generating code, but fixing it

Most professionals who have tested these tools agree on one point: AI accelerates specific tasks but does not replace judgment. "Sometimes I spend less time programming something myself than asking AI, given that it makes so many errors that you take longer to fix them," admits another. AI is useful for exploring approaches or generating ideas, but someone who understands the system still holds the wheel.

Some go further, arguing that the problem is not code quality, but the illusion that anyone can program without knowledge. "People who have never programmed think using a programming language is like translating a book from English to Spanish. And no, it is more like writing a book: you have to create a plot, research, develop characters, subplots, climax, and resolution," illustrates a participant. The metaphor is accurate: AI can draft sentences, but someone must know what story to tell.

The most optimistic calculation assumes a non-programmer can go from zero to an acceptable level just by sending written commands. The pessimistic scenario starts from the premise that this same user will be unable to detect the errors introduced by AI, and the cost of correcting them will fall on the few remaining experts. "Or maybe in two years you will need 10 people to fix the mess," points out another.

What happens with salaries and jobs in programming?

Despite the noise, sector salaries have not dropped. The influx of new professionals and the arrival of AI have not caused, for now, a collapse in compensation. "Salaries have not fallen despite the large number of people entering the sector + AI. When they drop, we will start talking about surplus workers," concludes a comment. Pressure on employment, for the moment, is a hypothesis, not data.

What seems clear is that companies will use AI to squeeze more out of their staff. "They will force office workers sitting in front of a computer to be much more efficient and produce more work, with the same salary, because they have an AI with which they can work faster," predicts another. This is the pattern already seen with the arrival of the internet in offices: higher productivity, same pay.

The business behind the announcement: funding rounds and hype

Behind the grandiose statements lies an obvious commercial interest. "What he means is that they extend him another funding round paid for by others, since he already knows what to do with the money, for everyone's good," ironizes a skeptic. The CEOs of major AI companies need to maintain the narrative to attract capital, and catastrophic or miraculous headlines are part of the marketing.

It is not the first time the end of a profession has been announced. "I used one of the first code generator programs when starting my career. We are talking about the year 2001. It generated COBOL pseudocode. And the tool then was only marginally worse than current AI tools," recalls a veteran. History repeats itself, but with a bigger budget.

Real impact: rapid prototypes, impossible maintenance

Where AI is showing results is in prototype generation and documentation. "I am using Claude Sonnet 3.7 to generate code and have advanced in tasks that would have taken me days in just a few hours," confirms a user. Another adds: "Right now I am using Claude Sonnet 3.7 to generate massive technical documentation for a client that would take me days to do; with Claude it takes hours." Productivity in bounded tasks is real.

The problem arrives with maintenance. If AI generates thousands of lines of code that no one fully understands, the cost of fixing a bug can skyrocket. "If there is no human behind it, it will make the same accounting program, the same app, and the same video game for all users. So the added value of using AI will be zero," warns another. Homogeneity is the enemy of competitive advantage.

The prediction no one dares to date

The discussion is not whether AI will transform programming, but when and at what pace. "It is not about guessing the future, it is about positioning yourself with the current, or at least rowing against it. One thing is what technology allows, and another is what society is capable of accepting at a given moment," reflects a participant. Adoption depends on legal, economic, and cultural factors that do not advance at the estimulante ilegal of language models.

Meanwhile, programmers keep typing code. "He said that last year too, and I still have to type crap," summarizes one. AI writes better every day, but the person signing the check still wants to know who fixes the mess when something fails. And that someone, for now, is human.