The Real Fight: Speed From AI Code Generation vs Trust In Open Source Quality
The current debate over AI code generation in open source software is about whether faster contributions from language models can coexist with the community’s need for trustworthy, well-understood code, reliable maintainers, and meaningful review and mentorship. That tension, not hype about machines replacing programmers, is what is driving new policies, new governance models, and new cultural norms in flagship open source projects. Maintainers are discovering that the cost of reviewing AI pull requests often outweighs the benefits, so they are prioritizing code review standards, contributor understanding, and long-term sustainability over short-term speed gains. The Godot game engine team has become the emblem of this backlash. They announced that they are rewriting their contribution policy to prohibit almost all use of AI from contributors after an overwhelming number of pull requests appeared to be AI-generated. Their blunt assessment — “AI cannot take responsibility, and we can’t trust heavy users of AI to understand their code enough to fix it” — should be read less as luddite panic and more as a defense of open source quality. When reviewing becomes a demoralizing slog through vibe-coded patches, maintainers burn out and the project suffers. That is the core risk AI is exposing.
Godot’s Hard Line: Banning "Vibe-Coded" AI Pull Requests To Protect Reviewers
Godot’s stance is deliberately harsh, and that is exactly why it matters. The team is reacting to a pattern: a flood of AI pull requests from contributors who do not understand what they are submitting and who cannot respond meaningfully to review feedback. Maintainers have described these AI pull requests as “demoralizing” wastes of time. In a culture where volunteer energy is the scarcest resource, demoralizing work is an existential threat. So Godot is rewriting its rules. New contributors with three or fewer merged pull requests must get explicit permission before submitting new features or significant refactors, a filter aimed squarely at vibe coders and autonomous agents. AI-authored garbage and agent-generated code now means an auto-ban from the project’s GitHub repository, and the prohibition extends to any substantial piece of code produced by AI. This is not just gatekeeping; it is a conscious decision to nurture contributors who understand the codebase and are willing to talk to maintainers to learn more. In other words, Godot is formalizing mentorship and stewardship as higher priorities than raw development speed.
Eve Online’s Carbon Engine: Openness, Security, And Declared LLM Use
While Godot slams the door on most AI code, Eve Online’s Carbon engine takes a more nuanced path: open source everything, but tighten governance around both security and LLM-written contributions. Fenris Creations decided two and a half to three years ago that there was “nothing really special” about their code and that the community would be better served by putting it on GitHub so more people could inspect and extend it. The tech behind a game with an in‑game trading volume estimated at more than $50 million per year is now broadly available, under permissive licenses, to anyone who wants to build on it. Openness, however, comes with responsibilities. The core tech team has been reserving time during its sprint process to review pull requests and handle governance. Security is “absolutely” a concern, and the pressure has forced more effort into review and architectural decisions. At the same time, Fenris is confronting AI coding head-on: contribution guidelines now include criteria for testing work and a requirement to disclose if a contributor has used an LLM. “We don’t mind you using an LLM, but you have to disclose it because we may subject it to different scrutiny than if it were not disclosed”. That single sentence captures a mature stance: AI code is allowed, but it earns extra skepticism and must meet tougher code review standards.
Linus Torvalds: Stop Panicking, AI Is Just Another Layer Above The Compiler
Against this backdrop of exhausting AI pull requests and meticulous governance, Linus Torvalds offers a badly needed reframing. While CEOs and investors make breathless claims that AI now writes ninety or even ninety‑five percent of their code — with one chief executive saying ninety percent could be written by AI within three to six months, and another predicting ninety‑five percent within five years — Torvalds responds with a simple observation: one hundred percent of that code is already written by compilers. His point is not pedantic; it is a reminder that automation has always been part of programming. “AI is a great tool, but it’s a tool,” he says. He argues that describing code as “written by AI” is the same category mistake as saying your code is “written by GCC.” Compilers translate human intent into machine instructions and nobody holds a press conference about that. AI code generation, in his view, is just another layer in the abstraction stack — human intent to AI‑generated source, then compilers, then assemblers, then machine code. AI may change how we write software but not the fundamentals of what programming is. This is exactly why panicked narratives about AI replacing programmers miss the mark, and why the real debate is about quality, responsibility, and how tools fit into long‑standing workflows.
Where Open Source Goes Next: Quality Control And Mentorship Over Hype
Put Godot, Carbon, and Torvalds together, and a clear direction emerges. AI tools like Cursor, Claude Code, and GitHub Copilot are changing day‑to‑day coding workflows, but open source communities are refusing to let the hype dictate their values. Godot’s maintainers explicitly say they need to ensure that reviewers feel their time is well spent. Fenris Creations frames its open source move as a way to gather interest so people invest time, effort, and money into contributions that lift all ships. Both projects are putting contribution guidelines, PR templates, and governance structures in place to sustain that culture over the long term. What comes next is not a rush to have AI write everything, but a consolidation of norms. Godot is updating contribution policies and enforcing stricter rules on AI code. Carbon’s team will create example test projects and continue to reserve sprint time for PR review. Fenris is even preparing to open source its own tools gateway for LLM interfaces after some hardening. The direction of travel is clear: AI code generation is welcome as a tool, disclosed and scrutinized, while human responsibility, mentoring contributors, and protecting open source quality remain non‑negotiable. That is a healthier future than any promise that ninety percent of our code will be “written by AI.”
