Software design is now cheap

AI makes coding cheap — devs debate miracle vs mess

TLDR: A viral essay says AI assistants make software design cheap, like 3D printing for code, without a quality hit. Commenters split between hype and hazard: boosters tout faster custom tools and fewer dependencies, while critics warn about long-term maintenance, security leaks, and LLMs struggling with real design.

The hot take of the day: a buzzy essay claims large language models (LLMs) have made software “design” cheap, like 3D printing did for workshop jigs — except without the quality penalty. The author argues AI can crank out one-off scripts and custom pieces so fast that the dusty catalog of standard parts (think endless libraries and frameworks) suddenly looks optional. Fewer dependencies, fewer glue layers, fewer headaches — at least in theory.

Cue the fireworks. Skeptics came in loud: themafia sneered that “vibe coders” are just “temporarily embarrassed billionaires,” warning that ditching standard parts means ditching proven maintenance paths. Security hawks like agentultra waved red flags about AI-made apps “leaking keys” and turning into unmanageable Franken-projects. Meanwhile, amluto split hairs, saying AI can grind through implementation but flops at design — it’ll ship code, just not good ideas. On the flip side, barishnamazov insisted the opposite: if you design clearly, the AI does grasp your concepts and fills in the details.

The thread turned into meme city: folks riffed on the essay’s “two nuclear plants for a porch light” with “AI that prints steel… and surprise bugs,” while others cheered the dream of a smaller “junk drawer” of dependencies. Bottom line: speed-drunks vs. battle-scarred maintainers — and nobody’s backing down.

Key Points

  • Software’s main cost is design; fabrication (compilation and copying) is cheap, leading to heavy use of standard components.
  • Physical manufacturing uses 3D printing to reduce fabrication costs for jigs and prototypes, but printed parts have quality limitations.
  • Software has analogous jigs (scripts, converters, test harnesses) that were often skipped due to high tooling costs before LLMs.
  • LLM-assisted coding reduces the cost of creating one-off tools and enables rapid generation, use, and disposal of custom code.
  • LLM-generated code lacks an inherent quality penalty; if correct, it matches hand-written code, increasing feasibility of custom solutions and emphasizing testing.

Hottest takes

“Vibe coders really embody the "temporarily embarrassed billionaire" mindset” — themafia
“LLMs are really quite bad at designing anything interesting” — amluto
“LLM will generate the understanding as well” — barishnamazov
Made with <3 by @siedrix and @shesho from CDMX. Powered by Forge&Hive.