If the generated implementation is not good, you're trading short-term "getting on with the job" and "real deadlines" for mid-to-long-term slowdown and missed deadlines.

In other words, it matters whether the AI is creating technical debt.

If you're creating technical debt, you're creating technical debt.

That has nothing to do with AI/LLMs.

If you can't understand what the tool spits out either; learn, throw it away, or get it to make something you can understand.

Do you want to clarify your original comment, then? I just read it again, and it really sounds like you're saying that asking to review AI-generated code is "disingenuous or naive".

I am talking about correctness, not style, coding isn't just about being able to show activity (code produced), but rather producing a system that is correctly performing the intended task

Yes, and frankly you should be spending time writing large integration tests correctly not microscopic tests that forgot how tools interact.

It's not about lines of code or quality it's about solving a problem. If the problem creates another problem then it's bad code. If it solves the problem without causing that then great. Move onto the next problem.

Same as pretending that vibe coding isn't producing tons of slop. "Just improve your prompt bro" doesn't work for most real codebases. The recent TEA app leak is a good example of vibe coding gone wrong, I wish I had as much copium as vibe coders to be blind to these things, as most of them clearly are like "it happened to them but surely won't happen to ME."

> The recent TEA app leak is a good example of vibe coding gone wrong

Weren't there 2 or 3 dating apps that were launched before the "vibecoding" craze that went extremely popular and got extremely hacked weeks/months in? I also distinctly remember a social network having firebase global tokens on the clientside, also a few years ago.

So that's an excuse for AI getting it wrong? It should know better if its so much better.

LLMs are not meant to be infallible it's meant to be faster.

Repeat after me, token prediction is not intelligence.

Not an excuse, no. I agree it should be better. And it will get better. Just pointing out that some mistakes were systematically happening before vibecoding became a thing.

We went from "this thing is a stochastic parrot that gives you poems and famous people styled text, but not much else" to "here's a fullstack app, it may have some security issues but otherwise it mainly works" in 2.5 years. People expect perfection, and move the goalposts. Give it a second. Learn what it can do today, adapt, prepare for what it can do tomorrow.

No one is moving the goalposts. There are a ton of people and companies trying to replace large swathes of workers with AI. So it's very reasonable to point out ways in which the AI's output does not measure up to that of those workers.

I thought the idea was that AI would make us collectively better off, not flood the zone with technical debt as if thousands of newly minted CS/bootcamp graduates were unleashed without any supervision.

LLMs are still stochastic parrots, though highly impressive and occasionally useful ones. LLMs are not going to solve problems like "what is the correct security model for this application given this use case".

AI might get there at some point, but it won't be solely based on LLMs.

> "what is the correct security model for this application given this use case".

Frankly I've seen LLMs answer better than people trained in security theatre so be very careful where you draw the line.

If you're trying to say they struggle with what they've not seen before. Yes, provided that what is new isn't within the phase space they've been trained over. Remember there's no photographs of cats riding dinosaurs but SD models can generate them.

Saying that they aren't worse than an incompetent human isn't a ringing endorsement.

I've heard this multiple times (Tea being an example of problems with vibe coding) but my understanding was that the Tea app issues well predated vibe coding.

I have experimented with vibe coding. With Claude Code I could produce a useful and usable small React/TS application, but it was hard to maintain and extend beyond a fairly low level of complexity. I totally agree that vibe coding (at the moment) is producing a lot of slop code, I just don't think Tea is an example of it from what I understand.