Dave Smith
Software developer who co-hosts Soft Skills Engineering, a weekly advice show for software engineers.
“We are walking into a tidal wave right now that is washing up on shore. And that tidal wave is AI-generated code. And I am convinced that we cannot oppose this tidal wave. We have to find a way to live with this volume, with this new way of writing code. And we cannot insist on the old way of carefully reviewing every line.”
Soft Skills Engineering · Hands-On Engineering Podcasts · October 2026
Answering the same listener, Smith said the complaint carried an expectation that code would still look like the carefully written, human-authored code of five years ago. He added that code which doesn't do what the spec says is bad engineering whether AI wrote it or not.
Transcript
Not only that, but it just won't be good. You know, it just won't be what you want.
Yeah. So you have an expectations problem on your team where you expect that even if they're using AI to produce it, it still feels like a human produced it. And so you rewrite it, right? You even said that in the question. I rewrite it, and now it reads like a human wrote it because that's what I expect. I expect it to have enough care and thought to indicate a human made this by hand. And they are not doing that. And I don't think they'll prompt their way to looks as if a human wrote it, if nothing else than just from the sheer level of output. So I think you have to decide: do I want to expect them to do this by hand or produce handwritten looking output or not? And that's separate from like, is the output any good? Like,
does it do what the spec says to do? Which this question asker says it does not. Yeah. And that's just bad engineering, regardless of whether it's AI or not, if that's the case. Yeah. I mean, by the way, I think you're right, Jameson, that this question asker is indeed, does indeed carry an expectation that the code will look like the code that we used to review five years ago. You know, carefully thought out, human-authored. But I don't know how else to say this, but we are walking into a tidal wave right now that is washing up on shore. And that tidal wave is AI-generated code. And I am convinced that we cannot oppose this tidal wave. We have to find a way to live with this volume, with this new way of writing code. And we cannot insist on the old way of carefully reviewing every line and having it feel like human, curated, carefully written, so that a human can read it. It's just, we just can't have that anymore, I think. Whether we like it or not. You know, it's just, I don't think we can have that. So I have to ask, how do we learn to live in this new world? I
don't know if I could clearly articulate what good AI planning output looks like. I feel like that's something we need to figure out if we're going to use AI to author not just the code, but these kind of meta documents. I don't know. I just quickly jumped to, well, obviously this was written by AI, and therefore it has all of these problems. And unless you rewrite it, I don't know how you get rid of the it was written by AI sense, even if you are able to make it good, you know? Like, so I think we have like a signal problem to show, no, no, no, like, yes, this was produced with AI, but also I have, I have worked on it enough with my own human brain to put my stamp of approval on the output.
Yeah, I think there's one really simple heuristic, which is how many prompts did it take to get the document to its current state? And if the answer is like less than, I don't know, 20, then you probably have a crap document when it comes to like design docs, planning docs, large ticket planning docs, things like that. Maybe, maybe 10 to 20, somewhere in that range. But that may be what needs to happen here. We need some kind of mechanism that says this was a one-prompt document.
Yeah, we need to like, there's some new crypto algorithm. I mean, cryptography, not cryptocurrency. Yep. Although I'm sure, sure, put an NFT on it, I
More from Soft Skills Engineering
