People keep asking when AI will replace engineers like it’s a binary thing. One day you’re employed, the next day you’re not. That’s not how any of this works.
We’ve heard this before. Low-code was supposed to do it. No-code was supposed to do it. Squarespace. Wix. Wordpress. Every abstraction layer that let non-engineers build things was going to be the end of us. And yet. More engineers than ever. Why?
Because the job was never typing code. The job is understanding what to build, why it breaks, and building novelty custom tailored to new problems. Every tool that makes building easier just means more people want things built. The demand scales with the capability.
What’s actually changing?
Engineering hasn’t become obsolete. Memorizing syntax has become less important for dev velocity. And I’ve never really valued that anyway. Happy to have less carpal tunnel personally.
We’ve been using GitHub Copilot for almost three years now. Wow. Before that we had intellisense, which also ate away at our ability to write code “by hand.” My muscle memory for specific syntax has probably degraded. But I haven’t noticed because I write so little code by hand in the last six months.
Does this matter? I can still identify architectural issues and code smells. I’ve definitely gotten better at intuiting when to abstract ahead of time. These skills have always been more interesting anyway. The boring mechanical stuff was never the point.
What makes an engineer valuable?
If your value as an engineer came from writing highly performant binary tree implementations in C++ at 145 wpm, yeah. You’re in trouble. But that was never the job.
The job is understanding systems and which systems to build. Knowing when something will break and how to minimize blast radius before it does. Seeing the abstraction that’ll save you six months of maintenance debt. Reading code and immediately sensing that something’s off. Those skills haven’t degraded. If anything, they matter more now because you can move faster through the mechanical parts.
Can AI do these things? Sometimes. Does it do them reliably? Not yet. Not without someone who knows what good looks like checking the work. And at what point will stakeholders not want someone in the loop who can actually validate? Probably never.
The validation question
Here’s my critical caveat. I’ve written and reviewed so much code that I don’t think I’ll ever lose my ability to validate outputs. I can read and reason about everything I produce.
Do I read everything I produce? Yes. Do I need to? Definitely not. Will I always read it all? I don’t know the future. Probably not.
Admittedly the primary reason I read it now is less to “check the code” and more to learn how I’m being interpreted. To understand how far I can trust the model next time. It’s calibration more than verification.
What about new engineers?
This is where it gets interesting. I have decades of writing code informing my ability to evaluate AI outputs. What about someone just starting?
Can you develop taste without writing the bad code yourself? Can you learn to spot code smells if you never smell them firsthand (ew what?)? I genuinely don’t know. Maybe the new path is different but equally valid. Maybe it’s worse. Maybe it’s actually better because you can iterate faster through more examples.
But I’d want to see it before I believed it.
So when?
When will AI replace engineers? Wrong question.
When will memorizing syntax stop being a meaningful skill? Already happened. When will understanding systems stop mattering? Not anytime soon. When will you need to actually understand what you’re shipping? Always.
The engineers who treat AI as a tool that still requires expertise will be fine. The ones who treat it as a replacement for understanding won’t be engineers anymore. They’ll be prompt typists. And you weren’t hiring them anyway, so relax.