Codeberg Just Banned AI-Generated Code. It Won’t Survive the Year.

You felt it before you read the announcement. That familiar twitch of someone moving the goalposts while you’re mid-sprint.

Codeberg — the open-source hosting platform many of us turned to as a GitHub alternative — just updated its Terms of Use to prohibit LLM-generated code. The pull request is live. The intent is clear: keep the platform human, keep the code clean, keep the AI slop out.

Noble? Sure. Nave? Absolutely.

When you build a wall against the ocean, you don’t stop the tide — you just drown on your own side of it.

Let’s talk about what’s actually happening here. The policy assumes a clean binary: this code was written by a human, this code was extruded by a machine. In 2024, maybe you could still pretend that line existed. A developer copies a chunk from ChatGPT, pastes it, maybe tweaks a variable name. The fingerprints are visible. The model leaves traces — verbose comments, oddly structured error handling, that unmistakable smell of confidence without understanding.

But that’s not where we’re heading. That’s not even where we are now.

The developer who asks Copilot to generate a function, reviews it, refactors it, renames variables, restructures the logic — is that AI code? The developer who writes pseudocode and lets the model fill in the implementation? The one who pair-programs with an assistant the way we used to pair-program with a colleague? Where exactly does Codeberg draw the line?

The dirty secret nobody wants to admit: within eighteen months, there will be no line to draw. AI-assisted code will simply be code.

This is the enforcement problem, and it’s fatal. Not in theory — in practice. Right now, Codeberg’s rule is enforceable because current AI output is detectable. It has tells. But models are getting better at mimicry, at matching a developer’s personal style, at producing code that reads like it came from a specific human with specific habits. The day that detection becomes unreliable — and we’re talking months, not years — this policy becomes a piece of theater.

What happens then? Two paths, and neither is good for Codeberg.

Path one: the rule becomes a dead letter. Everyone knows it’s unenforceable. Contributors submit AI-assisted code anyway. Moderators stop checking. The policy rots on the page like an old sign nobody reads. The platform keeps functioning, but its stated principles become a joke. Cynicism sets in.

Path two: the rule is enforced selectively, aggressively, and the community fractures. Purists police every PR. Contributors who use AI tools — which will be most of them — leave for platforms that don’t interrogate their workflow. The projects that remain are smaller, slower, and increasingly irrelevant. Codeberg becomes a museum for people who write code the old way, preserved in amber while the industry moves on.

You can’t gatekeep a paradigm shift. You can only choose whether you’re part of it or a footnote to it.

And let’s be honest about who this hurts most. It’s not the casual hobbyist who’ll just shrug and move to GitHub. It’s the developer who adopted AI tools early, who integrated them into a genuine workflow, who became more productive, more ambitious, more willing to tackle problems they’d never attempt alone. These are the builders you want in your community. These are the people who’ll leave first.

The top comment on the proposal already nails it: “Within a very short period of time all code will be AI code. What then?” It’s the right question, and Codeberg doesn’t have an answer.

Because there isn’t one. The trajectory is set. Models will improve. Code quality from AI will rise. Bug fixing, refactoring, optimization — the models will handle it. The distinction between “human code” and “AI code” will dissolve not because AI gets so good it’s indistinguishable, but because the workflow itself will fuse them so completely that separation becomes meaningless.

Think about it this way: nobody asks whether your code was written in Vim or VS Code. Nobody checks if you used autocomplete. Nobody audits whether you copied a pattern from Stack Overflow. We accepted those tools because they made us faster. AI assistants are the same thing, one order of magnitude further. The only difference is that the tool is now smart enough to make people uncomfortable.

Discomfort isn’t a principle. It’s a feeling. And feelings don’t make good policy.

Codeberg is reacting to a real problem — AI-generated slop flooding repositories, low-effort projects clogging search results, code that looks right but breaks in production. Those concerns are legitimate. But the solution isn’t a blanket ban that treats all AI involvement as contamination. That’s using a sledgehammer to address a quality problem that requires a scalpel.

The real conversation should be about quality gates, not origin gates. Does the code work? Is it maintained? Does the contributor understand what they’re submitting? Can they answer questions about it? These are the questions that matter. Whether the initial draft came from a human brain or a language model is increasingly irrelevant to whether the code deserves to exist on a platform.

If you’re a developer who uses AI tools — and statistically, you are — this policy should make you pause. Not because Codeberg is wrong to care about quality, but because they’ve chosen to define quality through provenance instead of outcome. And that choice, however well-intentioned, puts them on the wrong side of history.

The platforms that survive the next decade won’t be the ones that banned AI. They’ll be the ones that figured out how to make AI-assisted code better.

Codeberg has a choice to make. Walk back the ban, replace it with quality-focused guidelines, and stay relevant. Or dig in, enforce the unenforceable, and watch the community bleed out one contributor at a time.

The clock is already running. And it’s not on Codeberg’s side.