November 26, 2025
Cue the code wars
Vibe coding: What is it good for? Absolutely nothing
Old-school devs say AI 'vibes' break real apps; fans call it the future
TLDR: A fiery op‑ed says AI “vibe coding” is unpredictable and hard to maintain, turning prototypes into headaches. The comments erupt: pro‑AI users call the critique outdated and unfair, skeptics cheer the warning, and the split shows how teams are battling over whether AI should write code at all.
An opinion piece came in hot, declaring AI “vibe coding” good for flashy demos but bad for real-life software. It says the results change each time, tweaking prompts is a gamble, and you end up with code nobody understands. Even Linus Torvalds comparing it to typing programs from old magazines didn’t sway the writer, who invokes Edsger Dijkstra to warn about messy, unstructured code. The big claim: there’s no learning path here, just prototypes that turn into monsters the moment they look vaguely functional.
Comments lit up like a bug tracker at 5 p.m. on Friday. One camp calls the piece anti-AI slop, arguing non‑determinism doesn’t stop you from iterating and that LLMs (chatty code tools) give useful stuff “almost every time.” Another camp cheers the caution, saying corporate teams can’t maintain mystery code. A top quip: “New thing bad, old things good” became the unofficial meme, with replies like “OK Vibe Boomer” and “Schrödinger’s stacktrace: it works until QA looks.” Some want AI that teaches, not just spits code; others say we already have that if you use the tools right. Verdict? The vibe is divided, and very online. Expect more spicy threads as companies pick sides this year.
Key Points
- •The article describes “vibe coding” as AI-generated code from natural language prompts, similar in promise to low-code/no-code.
- •It argues vibe coding is non-deterministic, making iterative refinement unreliable and dependent on AI interpretation.
- •Maintaining AI-generated codebases is presented as problematic, especially as tools change, with few comparable production successes in low/no-code over decades.
- •The piece notes accessibility benefits, citing Linus Torvalds’ comparison to magazine-era BASIC, but recalls historical critiques of BASIC (including Dijkstra’s).
- •It concludes that vibe coding lacks a learning pathway to deep understanding and suggests AI should guide learning and environment setup rather than generate full code.