If AI writes the code, what are engineers for?

Many of us are asking the same question: will my job still exist in five years? Or even next year?

In March 2025, Dario Amodei predicted that AI would be writing 90% of code within three to six months. Programmers would still make design decisions and decide what to build. But he also expected AI to take over those tasks eventually.

AI systems have reached gold-medal performance in competitive programming, and companies are using them for software development. That’s real progress. It doesn’t tell us how much of engineering agents can take over, or how many engineers we’ll need.

For me, the change started with something much smaller.

What GPT-4 changed for me

Back in 2023, I was a data scientist with a strong software engineering background but little experience deploying services or models. I needed to build batch training and inference pipelines, so I was learning the infrastructure as I went.

I remember pasting Terraform snippets into GPT-4 and asking it to explain how things worked. It only saw what I pasted, not the full codebase, and the Terraform it generated often needed fixing. Still, it helped me understand what I was doing and get the work done faster.

I was still writing those modules by hand, but I think it would have taken me twice as long without GPT-4.

That was when I realized how useful LLMs could be as teachers. I could learn while doing the work. I still use them that way.

But what I needed to do myself in 2023 tells me little about what agents will need me for in ten years.

More code doesn’t settle the question

Writing code is only part of software engineering. We also have to work out what a system should do, how its parts fit together, and whether it actually works.

If generating code keeps getting cheaper, more projects may become worth building. That could mean more systems to maintain, more connections between them, and more decisions about what to change.

I expect engineers to keep helping with those decisions, using agents to do more of the work. Understanding why a system exists and what would count as a good result still seems valuable to me.

But more complexity doesn’t automatically mean more work for humans. Agents might get better at understanding systems, checking changes, and managing that complexity too. Even expensive agents could cost less than the human work they replace.

We could build more software and still need fewer engineers. Whether new projects make up for the work being automated is a separate question. The examples above don’t answer it.

My bet

I don’t think software engineers are about to disappear. I expect the work to keep changing as agents take on more of it.

But there’s a difference between a profession continuing to exist and today’s jobs being safe. I don’t know how many engineers we’ll need, or how much of the work we’ll do ourselves.

My bet is still that software engineering will be a job in ten years. It’s a bet, though. I could be wrong.