Godot vs AI: Why They Just Said NO!

Overview

Programming tools and AI coding assistants have made it dramatically easier to generate or modify large codebases and submit changes via pull requests—sometimes within seconds. However, the video argues this has created a major maintenance crisis in open-source projects:

• AI-generated pull requests flood maintainers and overwhelm the human review capacity.
• The resulting code is often hard for contributors to understand, may lack proper testing, and can include “hallucinated” outputs.

Key Points Raised

AI pull requests can be draining and demoralizing

The video highlights comments from Remi (head of the Godot Foundation maintenance) stating that AI-written pull requests have become increasingly draining. Maintainers must repeatedly second-guess submissions, including:

• Whether the code is human-authored
• Whether the author understands the changes
• Whether the tests are real
• Whether bugs came from AI mistakes versus human inexperience

Godot’s contributor model is being undermined

Godot’s model depends on welcoming new contributors and spending reviewer time to help them learn, fix issues, and eventually become trusted maintainers.

The video argues that LLM-driven contributions change this dynamic because AI-assisted contributors often don’t develop the same understanding or take ongoing responsibility for mistakes.

The review bottleneck is the core issue

Even if AI increases the number of incoming changes, approval and review remain human-only. Reviewers are already limited, so more AI-generated work adds noise without increasing meaningful contributor growth.

Godot’s Response: Stricter AI-Related Contribution Rules

The video says Godot has amended its contribution policies to address these concerns. Major changes include:

• No autonomous AI agent/vibe coding (auto-banned by GitHub)
• No AI use to generate substantial code
  • AI assistance must be limited to minor tasks (e.g., code completion, regex, find/replace)
• Mandatory disclosure if AI is used for code authorship
• No AI-generated text in communication with maintainers (respect principle)
  • Human-written original text can be translated if needed
• All pull requests must be reviewed and approved by a human before merging

Zig as a Parallel Example

The video cites the Zig programming language community as taking an even stricter approach (“No AI, period”) after seeing similar patterns.

Zig’s reasoning is framed as “contributor poker”: projects bet that early contributions pay off because contributors improve over time and become trusted reviewers. In Zig’s view, AI contributions rarely lead to that improvement and instead create background noise, such as:

• Long hallucinated PRs
• PRs that appear fine but were allegedly based on LLM outputs

Future Concern: “AI Auditing” Replacing Humans

The video briefly notes an ironic possibility: the endgame could become AI systems auditing other AI-generated code—described as a “terrifying” prospect.

Overall Conclusion

Because the primary bottleneck is human reviewers, AI-enabled mass submissions can break the contributor/reviewer system. The video presents this as the central justification for both Godot’s and Zig’s restrictive policies.

Presenters or Contributors

• Mike (presenter, “Game From Scratch”)
• Remi (Godot Foundation head responsible for maintaining the Godot codebase; quoted via a referenced post)