Will AI replace programmers? What the people building AI coding tools say

None of the people in these conversations says programmers are about to disappear. They disagree sharply about which part of the job stays human, and for how long. For anyone deciding whether to learn to code, all four sources point the same way: learn it, learn it deeply, and learn to work with the models, because the lasting value is in judgment, debugging and design, not in typing code out.

Dario Amodei, who runs Anthropic, expects programming to change faster than any other field. His reason is the closed loop: a model can write code, run it and check whether it works. As AI takes over most of the literal writing, he thinks human work moves toward system design, architecture and user experience, until eventually AI does those too. He cites a benchmark of real software tasks rising from about 3% to about 50% in a year.

The four founders of the Cursor code editor believe AI should amplify programmers, not replace them. Good engineering, they say, is made of countless micro-decisions and trade-offs, and handing them off loses control. Their tool removes predictable typing. They are candid about limits: agents are not yet very useful for most tasks, and frontier models are strikingly bad at finding bugs when simply asked. Reviewing AI output becomes a bigger part of the job. To young people their message is that "people's taste and creative ideas will be magnified."

Sam Altman of OpenAI expects humans to do a lot of programming in five to ten years, but in a very different shape, with some of it done entirely in natural language. He suggests asking what share of tasks AI can do, not how many jobs it replaces.

ThePrimeagen, a longtime engineer and teacher, is the most skeptical. He does not expect programmers to be replaced anytime soon. His worry is that leaning on AI leaves people who never build real skill. When he relied on an AI coding tool, he introduced more bugs. His rule: "There's no substitute for the hours."

They agree that programming is changing fast, that repetitive code goes first, that today's tools still fail in important ways and that learning to code is still worth it. They disagree on whether natural language replaces code. Altman thinks it may for some people, the Cursor team says not entirely, and ThePrimeagen suspects it would turn into another precise language. They also differ on how far to trust benchmarks, which the Cursor team says are contaminated and miss messy real-world work.

The article adds a caution about who is speaking. Three of the four sources sell AI coding products or the models behind them, so their views are informed but not neutral. A survey discussed in one conversation found that only about 2.7% of developers highly trust AI output. Nobody gives figures for how many programmers will be needed.

What to do with it: learn to code anyway, and properly. Put in the hours on fundamentals before leaning on autocomplete. Use AI as a co-pilot, not an autopilot, and practice reviewing code. Build something large enough that design choices matter. Understand code before changing it. Build debugging and testing habits, develop taste and judgment, and expect to keep re-learning your tools.