If leaning into the LLM-hype makes me a bad programmer, then so be it. I probably wasn’t that good to begin with.
It is reasonable to believe that the act of instructing computers has changed for good, at least to the same extent as when we moved from writing assembly to higher-level languages. The only worthwhile argument now is the degree to which these coding agents generate practically usable code (practically is doing a lot of work here).
Blanket statements like “LLMs suck, they’re useless,” and “we won’t need programmers,” are easy enough to make. Anyone who has actually used these tools would be quick to dismiss them or assume that the proponents of such gospels are also evangelists for a hidden agenda.
Yet that doesn’t stop us from being polarized into camps that add nothing of value to the discourse.
For anyone watching these conflicting takes from the outside, it helps to think of LLM-assisted programming as a spectrum. We already have empirical evidence of LLM-assisted (or even fully vibe-coded) software providing value to users and generating revenue. We also see plenty of cases where LLMs hallucinate libraries, cite non-existent research papers, or introduce unnecessary complexity that would have been obvious to a seasoned programmer.
Swearing by either extreme is mostly a form of convenience or perhaps reassurance in a fast-changing world. A better approach is to ignore the noise and use these tools, despite their imperfections, to build things that actually work.
Live and let build.