Will AI Replace Programmers?

A few weeks ago, Dario Amodei — the CEO of Anthropic, the company behind Claude — said something that sent Reddit into a spiral.

The thread had hundreds of replies. Some people said coding was dead. Others said Amodei was lying to hype his product. A 21-year-old computer science student asked whether learning to code was even worth it anymore. Nobody gave a straight answer.

So here’s ours.

The honest number

According to Oxford Martin School research, software developers sit at 48% automation risk — right in the middle. Not safe, not doomed. The grey zone that nobody finds particularly satisfying to sit in.

But that number was calculated in 2013, before GPT-4, before GitHub Copilot, before Claude Code. The world has moved. The honest answer in 2026 is that nobody knows exactly where that number sits today. What we do know is that it’s moving, and it’s moving fast.

What AI is already doing

This isn’t theoretical anymore. If you work in software, you’ve felt it.

AI writes boilerplate. It generates unit tests from specifications. It documents codebases, reviews pull requests, suggests fixes, autocompletes entire functions. Tools like GitHub Copilot, Cursor, and Claude Code have gone from novelty to daily infrastructure at most serious engineering teams in under three years. A developer in one Reddit thread described his current day as solo work that would previously have required a full team — refactors that four years ago nobody would have considered attempting, features shipped in days that used to take months.

That’s real. That’s happening now.

What AI is not doing

Here’s where it gets more complicated.

The engineers at Anthropic reviewing Claude-generated code are not passive observers. They are defining what gets built. They are catching the subtle architectural flaw three layers deep that will cause a problem six months from now. They are making the judgment calls that determine whether a system is actually good or just superficially functional.

One senior developer put it this way: it feels like being an engineering manager, except instead of managing a team of humans, you’re managing a team of code-generating agents. The typing is gone. The thinking is not.

And the thinking — the knowing what to build, the understanding why something is wrong, the translating of human needs into technical systems — that still requires years of experience to do well. Which brings us to the part nobody wants to say out loud.

The junior engineer problem

Anthropic’s research found something that should concern anyone starting a career in tech. Since ChatGPT launched in late 2022, hiring of workers aged 22 to 25 into AI-exposed occupations has dropped by around 14%. These aren’t people being fired. They’re people who aren’t being hired in the first place.

The entry-level work — the implementation, the boilerplate, the first drafts — is being automated. And that work was how people learned. You spent two years doing the boring stuff and came out the other side understanding how systems actually behave. If that pipeline disappears, the senior engineers of 2030 have to come from somewhere. Nobody has a good answer to that yet.

We wrote about this in more depth in Even the People Building AI Are Being Replaced By It — which is worth reading if you’re in this industry.

So — will AI replace programmers?

Not exactly. But not never, either.

What’s happening is more specific than replacement. AI is replacing the output of programming — the code itself — faster than it’s replacing the judgment that determines what code should exist. For now, you still need experienced humans to direct it, review it, and catch what it gets wrong.

But the number of humans you need for that is much smaller than the number who used to write the code. Where you needed 100 programmers, you might now need 10 very good ones and a lot of AI. The 10 will be fine — probably better paid than ever. The other 90, and the next generation trying to become them, that’s the real question.

What this means if you’re a developer today

If you have experience and deep systems knowledge, you are probably more valuable right now than you’ve ever been. The tools make you faster. The judgment you’ve built over years is exactly what’s needed to direct them well.

If you’re just starting out, the path is harder than it was five years ago. The junior roles that used to exist as training grounds are shrinking. You need to be better, faster, and you need to understand AI tooling from day one — not as a novelty but as the primary way work gets done.

And if you’re considering a career in software for the first time: learn to think like an engineer, not just to write code. The code is increasingly the easy part.

Check your specific role

The risk varies enormously within software. A data-entry programmer and a distributed systems architect are both “programmers” — but their exposure to AI displacement is completely different. The Oxford dataset breaks this down by role, and the spread is wider than most people expect.

You can see the full software developer risk assessment on this site — including what tasks AI is most likely to absorb and what genuinely keeps developers employed. Or browse all tech roles and their risk scores to see where your particular specialism sits.