I was having a discussion with a friend recently about “no engineers” and “yes engineers.” His argument was that developers used to be “no engineers.” You would say no because software was expensive to build. And now that AI can build things so quickly, maybe engineers should become “yes engineers” instead. Say yes, let the code get written, see what happens, change it later.
I actually agree with part of this. AI should make us much more willing to experiment and much less precious about code. But I think the reason engineers used to say no was slightly different. It wasn’t just that building software took time. It was that every piece of software you shipped created something else you had to own and maintain.
Code being cheap to produce doesn’t mean code is cheap to own. Yes, AI helps with ownership too. It can help familiarize yourself with code, refactor and debug. But it doesn’t get paged at 3am when something breaks, deal with data loss, or handle a security incident. It also doesn’t have the context of why the system ended up the way it did.
That’s the part of engineering I don’t think AI changes. Someone still has to own the thing after it’s built.
I think there are basically two kinds of code. There’s disposable code: prototypes, experiments, one-off scripts, things you’re genuinely happy to throw away. With that stuff, I’d happily let AI go wild. Generate all of it without understanding, who cares.
Then there’s code that’s part of the product. For that, I don’t really care if AI wrote the code or a human did, what matters is whether you understand what you’re responsible for: how the system works, why you made the important decisions, where it can fail, and what happens when it does. You don’t need to understand every line, but you should be able to defend the system you shipped.
Maybe that’s the real shift AI is creating. We should probably say yes much more often. We can afford to try things that would have been too expensive to try before. But we also need to get better at recognising when something has gone from “who cares, it’s just code” to “this is now our system.”
The goal isn’t to eliminate the slop. It’s to get really good at knowing when to slop, and when to engineer.