3 Min Read

Logo and background image via Godot

Godot veteran Rémi Verschelde has stressed the free, open-source game engine is not being "vibe coded" after some users expressed concern over its current generative AI policy.

As it stands, the guidelines covering pull requests (PR) discourage generative AI usage and prohibit contributions made entirely using tech such as ChatGPT, Claude, Grok, and more. Yet, use of the divisive technology is permitted in some instances, including for translation purposes and single-line code completion. 

"We acknowledge that AI can be useful, but we are convinced that human effort results in better and more relevant contributions," read the current guidelines. "If you do use AI, we expect that you put in effort to proofread and improve anything it generates, and that you disclose what you used AI for. As maintainers spend significant time reviewing your code, please make sure that the code you submit is well-tested and functional. Please be respectful of their time and only submit something you have put thought and effort into." 

Related:French union STJV to commence national strike outside Quantic Dream

Verschelde—who has spent almost a decade working on the engine and co-founded major Godot backer W4 Games—has moved to provide additional context and claimed the Godot foundation and maintainers remain "fairly critical" of generative AI usage.

"I see some concerns raised about Godot's policy on contributions using AI. TL;DR: Godot isn't vibe-coded. As a whole, the Godot foundation and maintainers are fairly critical of AI usage, and don't make nor accept vibe-coded contributions. Godot doesn't include AI features nor intends to add any," he said in a thread on Bluesky. 

"Our policy tolerates some AI assistance in the contribution process ('some' is load-bearing: debugging, searching for information, surgical changes to existing code).I know it's already crossing the line for a number of people, for valid reasons, and I fully respect that position." 

In February, Verschelde said "AI slop" pull requests had become overwhelming for those attempting to maintain the engine. To address that issue, he explained Godot is preparing to roll out a more detailed generative AI policy that will be published soon. "It's a bit more restrictive, but not strictly a blanket ban," added Verschelde. 

He also suggested the idea that Godot is now "full of slop" is a wild misconception. 

"Any slop PR is automatically rejected, as simple as that," Verschelde continued. "All other PRs are submitted by humans who are expected to take responsibility over the code they submit, and they undergo a PR review by expert maintainers to ensure quality. They need to understand it, test it, be confident in the implementation, be willing to help maintain it if it’s merged.

Related:'Larger budgets and longer schedules do not lead to better games:' Thank You For Your Application recoups dev costs in 22 hours

"Likewise, they need to be able to assert copyright over it. That precludes any substantial vibe-coding. If a PR has an AI disclosure, or signs of undisclosed AI usage, it gets scrutinized more as we mechanically trust it less (especially from new contributors)." 

Verschelde explained Godot maintainers merged around 47 pull requests with AI disclosures out of 3,700 over the past two release cycles—so around 1.27 percent. That included some submitted by Verschelde himself, who has been experimenting with the technology to better understand its impact.

"I’m well aware that AI is a sensitive topic, and that some of you may be upset to learn that Godot even tolerates some of it," said Verschelde. "I believe we have a fairly pragmatic stance that keeps AI at arm’s length, based on our experience maintaining Godot for 12 years. I know that won’t satisfy everyone."

About the Author