"The LLM Critics Are Right. I Use LLMs Anyway"

What's this about?

A developer's essay holding two truths at once: the standard criticisms of LLMs (unreliability, skill erosion, homogenized output) are largely *correct* — and the productivity gains are real enough that he uses them anyway. Core claim: "LLMs amplify what you already have — if you have thoughts, they come out sharper and faster."

What exactly is the "skill atrophy" worry?

If an AI does your lower-level work (writing the code, drafting the prose), the mental muscles for that work weaken — like navigation sense after years of GPS. The sharpest version in the thread: your high-level judgment is *built from* lower-level experience, so outsourcing the low level may eventually hollow out the high level too. Even LLMs demonstrate this — you can't delete basic facts from a model and keep only "higher-level thoughts."

What's the community saying?

1. Atrophy is real, but is it bad? — the abacus/calculator precedent: we lost mental arithmetic and mostly didn't suffer; is coding the same kind of skill or a load-bearing one?
2. The homogenization fear — "after 5 years, every LLM power user's thought profile will be an LLM-derived carbon copy of each other."
3. No single middle ground — the best-received framing: the right human/AI split is a personal gradient descent that depends on who you are and what stage you're at.
4. Undeniable speedup — even skeptics concede gains in familiar areas and faster on-ramps into new ones; denying it "makes you look uninformed."