"AI Can Make You Suck Faster Too"

What's this about?

An argument that AI amplifies bad practice as readily as good — and a thread that finds a real hole in the methodology.

What is the methodological objection?

The post reaches conclusions about LLM coding ability while testing only DeepSeek. Whatever you think of the thesis, evaluating frontier capability without frontier models is a straightforward problem, and the thread says so. The counter-argument is also fair: "you used the wrong model, try a different one" is the standard unfalsifiable defence, and the actual productivity research shows minimal gains at best. Both objections are correct, which is why the thread does not resolve.

What's the community saying?

1. The line everyone quotes, from the pro-thesis side: "Pre-AI, the limiter on my bad code was how much code I could type in a day. Now I can 10x my output and 10x my bad code."
2. The optimistic reply — whoever gives a bad coder resources will eventually stop — gets demolished three separate ways: the buyer sees working software, not a mess; at a decently sized company delivering features is all that matters and the author is promoted twice before cleanup starts; and you are underestimating how little most executives care.
3. The methodology objection about DeepSeek, and the counter about what productivity research actually shows.
4. It is the complement to yesterday's "AI-written code is still your code" — same problem, angrier.