YouTuber Theo, founder of T3, gave thousands to Codeberg. He saw them vote in AI restrictions and called it "spite-driven" and a "cult."
He is mistaken. It wasn't about purity, it was about setting a boundary.
What actually happened
Codeberg e.V. is a non-profit Git forge located in Berlin. You can see it as a European version of GitHub, however, it is operated based on a democratic membership structure.
That is an important point. Members do not receive policies in a top-down manner. They actually have the opportunity to vote on them.
Therefore, a 14-day asynchronous vote was held on the proposal in the period from July 9 to July 22, 2026. The vote was closed by the admins Bastian Greshake Tzovaras, Otto Richter, and William Zijl and the results were announced on the following day.
The proposal was approved by 358 votes in favor to 144 against, with 14 abstentions, for a total of 516 votes cast. That's 516 votes cast, roughly 71% approval, with about half of active members turning out.
The new Terms of Use state it directly: "Do not upload creations that are primarily composed of code written by 'generative AI'-tools (or services like Claude, OpenAI Codex)."
Why "cult" is the wrong word
A cult doesn't conduct an open vote and then lose 144 of them. A cult doesn't accept dissent this openly.
Codeberg did the most sensible thing here. They requested, they quantified, and they released the numbers of the losers as well as the winners.
Critics often overlook the fact that Codeberg explicitly stated that they would never resort to using automated content analysis to delete large numbers of repositories. Enforcement is done on a case-by-case basis, following the common sense of the community.
Small projects, experiments, and repositories that were created before LLMs are not affected. That's not zealotry. That's a maintainer team trying to protect their review bandwidth without becoming code police. 🤷
This is about review cost, not vibes
The real argument isn't "AI bad." It's who pays for the flood.
A paper published on arXiv on July 31, 2026 discussed open-source governance, and in the paper, a Matplotlib maintainer was quoted saying that AI "changes the cost balance between generating and reviewing code."
One more time. It became inexpensive to create. But it remained expensive to review.
Every PR that can't be automatically reviewed goes to a human volunteer. When you can spray a hundred plausible-looking commits in an afternoon, the bottleneck moves entirely onto the people saying yes or no.
Codeberg also feels the pinch in the real world. For instance, a storage drive they previously paid €700 for now costs about €3,700, and they attribute this to AI datacenter expansions distorting the prices of hardware.
It means that the same machines causing a data flood are also driving up the cost of the disks used to store that data. Isn't that a wild loop?
Boundaries attract, not repel
If these were considered behaviors of a cult, people would run away as fast as they could. But instead, it was quite the opposite.
The Zig programming language community moved its infrastructure over to Codeberg. Why? Well, GitHub was basically forcing AI-assisted entry points onto contributors, and Zig hoped to encounter fewer violations of its strict no-AI policy over on Codeberg.
Communities have the autonomy to reject contributions that are beyond their capacity to evaluate. This is what governance is all about. It's similar to a linting rule, CLA, or a code of conduct.
You may have a different opinion about where exactly Codeberg decided to draw the line. I myself am not entirely certain I would have drawn the line there.
However, having a different boundary in mind does not justify claiming that the one that was drawn is illegitimate.
The takeaway
Theo provided funding to an organization based on democracy, but became upset when the democratic decision was not in his favor.
Establishing a contribution limit is not being overly selective. It's a team of maintainers determining what they can responsibly handle. And they ask for feedback from everyone beforehand. âś…
I'd like to know what you think about this: How can we differentiate between a healthy community boundary and real gatekeeping, and what decision would you have made on July 22?
Top comments (2)
In practice, how do you distinguish between code “primarily” written by AI and code that AI merely helped refactor or debug? Since there’s no automated analysis—only a case-by-case evaluation—doesn’t that effectively amount to a subjective decision by a single maintainer? Doesn’t this create a risk of inconsistency—that the same type of code might be accepted one time and rejected another, depending on who happens to be reviewing it?"
I think review cost is the clean frame. Cheap code generation changes one side of the ledger, but it does not make review capacity cheap. A forge has to decide whether it is mostly a neutral storage layer or a community with a scarce maintainer budget. The second version can set boundaries without turning the policy into a moral claim about every tool.