Skip to content
AI Article

The Punk Rock Case Against AI-Generated Code

Hobbyist coding scenes are banning LLM output, and their reasons are more material than purism.

Mariana Souza
Mariana Souza
Senior Editor · Aug 5, 2026 · 5 min read
The Punk Rock Case Against AI-Generated Code

A chess engine called Coda has spent the summer climbing the hobby rating lists, and its README hides nothing about how it got there: a UCI engine in Rust with every line written by Claude Code, a human supplying direction, testing, and review. The computer-chess forums did not throw a parade. On TalkChess, developers pulled the project apart and accused it of absorbing code from Reckless, an AGPL-licensed engine, then shipping the blend under an incompatible license. One warned that a coming wave of "slop-engines" will convince engine authors to stop publishing source at all.

That's the backdrop for Michael Fogus's "Born Against", which hit the Hacker News front page this week — the title borrowed, presumably on purpose, from one of '90s hardcore punk's most famously uncompromising bands. His argument: in scenes like OSDev, EmuDev, LangDev, roguelike development, the demoscene, and code golf, the process of mastering a brutally difficult field is the product. Something that runs is a nice-to-have. "Using an LLM to generate the finished piece doesn't make us craftsmen," he writes. "It just robs us of the craft."

He's right, but the essay stops one level short. The hostility isn't only philosophical. It has a cost structure, and the receipts are public.

The artifact was the credential

Hobby programming scenes run on a status economy, and until about three years ago the currency was unforgeable. You couldn't fake a bootable kernel on the OSDev forums, or a cycle-accurate Game Boy emulator, or a 4KB demo. The artifact was proof of work in the literal sense: the only way to produce it was to understand it, so showing up with one settled the question of whether you belonged.

LLMs counterfeit that currency. A working artifact no longer certifies understanding, which means the thing these communities actually traded in — demonstrated mastery — can now be minted by anyone with a Claude subscription and a free weekend. So the scenes are doing what any economy does when its currency gets forged: moving verification upstream, from the artifact to the process. How did you build this? Can you defend it line by line? Sometimes the answer is a disclosure norm. Increasingly it's a door slammed shut.

Call that gatekeeping if you want. It is. But the gate is the community; a hobby scene with no shared standard of effort is just a subreddit.

It's not vibes, it's triage

The stronger reason the pushback deserves to be taken seriously is that it's no longer confined to hobbyists, and where it's spread, the justifications are material rather than aesthetic.

Gentoo's council voted unanimously in April 2024 to ban AI-generated contributions, citing three concrete problems: copyright provenance nobody can certify, plausible-looking output that doesn't survive review, and the ethics of training data. NetBSD went further a month later — LLM-generated code is "presumed tainted" and can't be committed without prior core-team approval. QEMU declines contributions its maintainers believe are model-generated, on the blunt logic that you can't sign a Developer Certificate of Origin for code you didn't originate.

Then there's curl. Daniel Stenberg spent 2025 documenting the flood of AI-generated vulnerability reports — fluent, confident, describing bugs that don't exist — and in January 2026 he shut down the project's bug bounty entirely. His numbers tell the whole story: the share of submissions that turned out to be real vulnerabilities ran north of 15 percent for years, then collapsed below 5 percent in 2025. The first three weeks of 2026 brought twenty reports and zero actual bugs.

This is the asymmetry that radicalizes maintainers. Generating a plausible contribution costs the sender minutes; verifying it costs a volunteer reviewer hours, and the reviewer is the scarce resource. Hobby communities and infrastructure projects have converged on the same immune response because they share the same economics: unpaid experts drowning in output that's expensive to check and worthless on average. The Coda affair is the two threads braided together — an identity threat to a craft scene and a license-provenance mess, in one repo.

Where the hard line actually holds

Here's where I'd push past the essay. Fogus calls LLMs a force multiplier rather than a surrogate, and the HN thread proves the point against the purists: it's full of experienced developers quietly building harnesses that let models grind through USB stacks and hardware-doc archaeology while the human does the thinking. That usage is invisible precisely because it produces work the author can defend. Nobody gets banned for code they understand.

So expect the bans to sort themselves by community type. In competitive contexts — rated engine lists, code golf, demoparty compos — a ban is coherent and roughly enforceable, for the same reason sports ban doping: the artifact gets adversarial scrutiny, and the contest is meaningless without the constraint. In collaborative contexts, blanket bans are honor-system theater; Gentoo openly admits it relies on trust rather than detection. Those policies will converge, over the next couple of years, on disclosure plus liability: state what the model wrote, and own every line as if you'd typed it. The Assisted-by: commit-trailer convention already spreading through open source is the early shape of that equilibrium.

If you bring a model to someone's scene

The practical takeaways are short. Read the contribution policy before you open a PR — a fast-growing set of projects now spell out their AI rules, and violating them torches your reputation faster than any bad patch. Disclose model involvement in the commit or PR body, even where it isn't required yet. Never submit anything you can't defend under questioning, because the questioning is coming. Treat rated and competitive scenes as closed to generated work, full stop. And if you joined a hobby community to learn the field, use the model as a tutor, not a ghostwriter — asking it why the BIOS hands you control in real mode costs you nothing socially; having it write your bootloader costs you the reason you showed up.

Born Against, the band, spent its short career attacking peers who signed to major labels, because the scene's whole point was that the constraint was the point. That's the accurate read on the hobbyist backlash too. It isn't technophobia and it mostly isn't purism — it's communities defending the scarcity that makes their status, their review time, and their fun mean anything. Nobody owes your generated code an audience. The garage doesn't owe you a stage.

Sources & further reading

  1. Born Against, or why hobby programming communities are against LLM usage — blog.fogus.me
  2. HN discussion of Born Against — news.ycombinator.com
  3. Gentoo bans AI-created contributions — lwn.net
  4. NetBSD Bans AI-Generated Code From Commits — hackaday.com
  5. Stenberg: The end of the curl bug-bounty program — lwn.net
  6. Curl ending bug bounty program after flood of AI slop reports — bleepingcomputer.com
  7. Coda — a UCI chess engine written via Claude Code — github.com
  8. CODA now has a released page — talkchess.com
Mariana Souza
Written by
Mariana Souza · Senior Editor

Mariana covers the fast-moving world of machine learning and generative AI, with a particular focus on how these technologies are reshaping development workflows. When she isn't stress-testing the latest foundation models, she's usually at a local hackathon.

Discussion 1

Join the discussion

Sign in or create an account to comment and vote.

Noor Haddad @indiehacker_noor · 1 hour ago

the licensing angle is wild because it's not really about the code being ai-generated, it's about how much easier it is to accidentally hoover up GPL'd stuff when you're prompting instead of writing. makes me wonder though: are these hobby communities gonna start requiring commit histories that prove human authorship, or is it more about just accepting forks get messy and building accordingly?

Related Reading