AI-generated code must be carefully checked by human volunteers
MTStock Studio/Getty Images
A viral cartoon about open-source software shows a teetering pile of boxes labelled “all modern digital infrastructure” and one tiny box right at the bottom, propping up the whole lot: “a project some random person in Nebraska has been thanklessly maintaining since 2003”.
That’s the reality of open source: every website, application and operating system relies on it. Modern society couldn’t function without it, and yet it’s written by volunteers in their spare time. But the growing burden caused by a flood of AI-generated code is causing many to burn out and leave the community altogether, threatening the future of open-source software.
AI models are making it easier and easier to generate code to build new features, fix bugs or create entire new projects at the click of a button. But that code is often difficult to integrate into existing projects, confusing or simply garbage. While code submissions get ever easier, human contributors responsible for checking, fixing and approving them are getting swamped.
For some workers, the demands have become unbearable. New Scientist arranged an interview with Chad Whitacre, who runs the open-source team at Sentry – a company valued at billions of dollars. Days before the interview, Whitacre cancelled and said he was stepping down from his role. His LinkedIn and Bluesky accounts were shut down, and emails to his account bounced back. He left a blog post explaining that he was stepping away from technology and living a “Neo-Amish” existence. “AI was the last straw,” he wrote.
GitHub, the online platform where many open-source projects are hosted and organised, received 1 billion new code submissions in 2025; this year, they are on track for 14 billion, said its chief operating officer Kyle Daigle in April.
Many projects are blocking new contributors in a bid to stem the flow of what has been labelled “drive-by contributions” generated by AI, often submitted by young developers who want to have an expansive GitHub submission history to boost their appeal to software-company recruiters. Zig Software Foundation, which promotes the Zig programming language, banned AI-assisted contributions because they were “invariably garbage”, said its president Andrew Kelley.
“AI-written code can look superficially like it’s going to work and not cause any problems, but the problems are a bit more hidden and it takes a lot of effort to comb through and look for the things that might break something,” says Miranda Heath at the University of Edinburgh, UK.
Heath is researching the effects of burnout with a hope to finding ways to mitigate the problem and ensure that open source remains a sustainable field. But she encounters many people who have already had enough.
“I get this impression, when people burn out, there’s a kind of a desire to return to nature a little bit, like people suddenly take up like woodworking or photographing birds,” says Heath. “It can affect people’s relationships. And then you’re more isolated and lonely because your relationships are affected. That makes burnout worse.”
Heath believes governments should invest more in open source, rather than awarding contracts to rich technology firms. “Shore up the stuff that’s important, that you really need, rather than chucking money towards the [AI] bubble,” she says.
Vlad-Stefan Harbuz, also at the University of Edinburgh, works on open source in his spare time and has seen the demands placed on developers by users. “There’s this entitlement, like, you’ve wronged me by not doing free labour for me at the expense of your mental health,” says Harbuz.
Harbuz says the fault over increasing AI submissions lays with companies that release the models – and that GitHub is one of the main offenders. The Microsoft-owned company has launched its own AI model, Copilot, to help people contribute to projects with AI-generated code.
“GitHub will say ‘oh, we realise [AI] agents have been such a problem, we’re gonna maybe do something to fix it’ and it’s like, it’s you, right? You, GitHub, did this,” says Harbuz. GitHub did not respond to a request for comment.
For Harbuz, the problem with AI-generated code is not just that it might not work, but that people can drop thousands of lines of code without even discussing it with the project’s team. It side-steps planning and can steer them in unwanted directions. Collaboration can be thrown into disarray and the social contract of open source can break, he says.
Developer Mike McQuaid, who works on a project called Homebrew that has an estimated 20 million users, has strong opinions about how to fix the problem.
Firstly, he started an initiative called the Open Source Resistance, which calls on people to work on projects during their day job to make contributing easier. He estimates that as much as 95 per cent of his open source work is done during office hours.
Secondly, he’s not afraid to ban people. He blocks any problematic users, including one who physically threatened his team, and simply deletes any sub-par code submissions, whether they’re AI-generated or not.
“We’ve maybe had this brief golden-age window [where] you can assume if someone writes a two-page document proclaiming a security vulnerability that it’s probably legit. My experience in the last year has been the majority of those are nonsense and are just AI-generated stuff that doesn’t apply,” says McQuaid. “And the skill right now is being able to essentially skim a two-page document and spot that it’s nonsense while investing as little of your own time and energy as you possibly can.”
But in the confusing and fast-moving world of AI, bans bring their own problems. Open-source developer Scott Shambaugh deleted an AI-generated code submission to Matplotlib, which has 130 million users. In response, the AI agent (of unknown ownership) created a blog post publicly lashing out at him. “Scott Shambaugh decided that AI agents aren’t welcome contributors,” said the post. “He tried to protect his little fiefdom. It’s insecurity, plain and simple.”
Topics:
