If you've been watching the endless flamewars over AI in software development and wondering why nobody can have a reasonable conversation about it, you're not alone. A new piece making the rounds on Hacker News offers what its author calls an "uncharitable taxonomy" of the AI discourse—and it's genuinely refreshing to see someone cut through the noise with actual categories instead of tribal warfare.

The Never Clankers and Their Sinking House

The first group identified are the "Never Clankers"—developers so opposed to LLMs that they advocate for excluding any contributor who has ever touched one from their projects. The author doesn't pull punches: "That position is a house on a broken foundation that is sinking in to the swamp." They point out that respected developers like Linus Torvalds and Fabrice Bellard have tested these models and found the larger ones useful enough for low-level work to admit they're sometimes practical. With AI integration becoming essentially inevitable across the software ecosystem, this purist approach is destined for irrelevance—noting that "the amount of software that is going to have no interaction with anyone who interacted with Claude is asymptotically zero."

Vibe Coders: Writing Bad Code Faster

On the opposite end sits what the author calls "Vibe Coders"—developers who don't even read the code anymore, just copy error messages back into prompts and ship whatever comes out. These are either people terrible at coding whose skills have atrophied from LLM overuse, or those who've figured out how to play corporate politics by letting AI write garbage while they collect paychecks. The author identifies this as a real problem: "These are the people that managers in love with artificial idiocy drool over—and is apparently empowering to do even more of it through agent fleets." Projects like MemVID are cited as examples where documentation is "heinously out of date all over the website and software," with policies likely changed by agents at the prompt level and never verified.

The Utility Minions Who Are Actually Fine

The third category might surprise you: "Utility Minions"—unskilled users who employ LLMs to create one-off scripts for personal needs. Someone using Google Gemini to write a script that reboots their routers, code that's absolute slop and probably nobody understands. But the author thinks this is actually okay: "10$ in token fees just gave them something they were otherwise completely unable to obtain." These users tend to be honest about what they're doing. The piece suggests they'd be even better served by bringing back 1990s-era power user tools like Warcraft 3 trigger editors or Macromedia Fusion event sheets—visual ways for non-programmers to build working systems without relying on black-box AI output.

The Demings: A Sustainable Middle Path

The most interesting category is labeled "The Demings"—named after W. Edwards Deming and his 100% inspection philosophy. These developers treat LLMs as just another tool in the box while imposing strict accountability: you're responsible for every line of code you ship, regardless of how it was generated. "You personally are signing off on releasing this code with your name stapled to it," the author notes. This approach is presented as the only sustainable one because it doesn't depend on any particular technology staying popular. Rules about refusing large undiscussed changes? Just good project hygiene, whether AI touched your codebase or not.

The Artificial Idiot Managers

The final category gets the harshest treatment: "Artificial Idiot Managers"—executives obsessed with replacing human workers entirely, not just reducing tedium. Citing Salesforce CEO Marc Benioff's claim that CEOs are "the last generation to manage only humans," the author argues this fundamentally misunderstands how industrial revolutions actually work. A sewing machine succeeded because it did a more consistent job faster and workshops could clear backlogs with them—"The leverage comes as a benny for the benefit." AI isn't yet shipping that benefit; executives just want to rush to the rewards first.

Key Takeaways

  • Anti-AI purism is doomed when respected developers like Torvalds openly acknowledge LLMs have legitimate uses
  • Vibe Coders represent real risk: LLM-assisted code without human review is "writing bad code faster"
  • Utility-level AI use by non-developers for personal tools is honest and probably fine
  • Full accountability for shipped code regardless of generation method is the sustainable approach
  • Managerial enthusiasm for replacing humans precedes any actual deliverable value from AI

The Bottom Line

This taxonomy won't win any popularity contests—the author admits it's "uncharitable." But it captures something true about the current moment: we're caught between purist rejection and reckless adoption, with both sides missing what should be obvious. Japan is already repealing privacy laws to appease AI scrapers while companies deploy LLMs regardless of fitness. Meanwhile, a handful of pragmatic people are actually getting useful work done and planning to stick around long after the hype cycle collapses under its own weight.