For open source software, AI writing tools are a mixed blessing

A world that uses increasingly powerful AI tools is a place where software development becomes cheaper – or so the thinking goes – leaving little room for traditional software companies. As one expert report it said, “vibe coding allows startups to replicate the look of complex SaaS platforms.”

Look at the hand twisting and notifications that the software industry is over.

Open source software projects that use coders for long-standing problems should be the first to benefit from cheaper code time. But that equation is not stable. In fact, the impact of AI writing tools on open source software has been very mixed.

AI writing tools have created as many problems as they have solved, according to industry experts. The ease of use and availability of AI coding tools has contributed to the proliferation of malicious code that has threatened to overwhelm projects. Innovating is easier than ever, but maintaining it is difficult and threatens to improve software development.

The result is a much more complex matter than the amount of simple programs. Perhaps, the predicted, imminent death of the programmer in the new era of AI is not yet over.

Quality vs quantity

Across the board, projects with open source codebases are seeing a decline in the number of submissions, likely due to AI tools lowering barriers to entry.

“For people who are small in the VLC codebase, the kind of requests we see are difficult,” Jean-Baptiste Kempf, CEO of the VideoLan Organization that manages VLC, said in a recent interview.

Kempf is still optimistic about AI writing tools but says they are best “for experienced developers.”

There have been similar problems with Blender, a 3D rendering tool that has been maintained as open source since 2002. The head of the Blender Foundation Franceso Siddi said that LLM-supported contributions often “wasted the reviewers’ time and distracted their attention.” Blender is still developing an official process for AI tools, but Siddi said it is “not mandated or recommended to contributors or developers.”

The flood of integration requests has gotten so bad that open source developers are creating new tools to manage it.

Earlier this month, developer Mitchell Hashimoto introduced a method that would limit GitHub’s offerings to “authorized” users, effectively closing the open source code. As Mr. Hashimoto said, “AI removed the natural barrier to entry that allowed OSS projects to rely on randomness.”

The same thing has happened in nutrition programs, which give outside researchers an open door to comment on safety issues. Open source data transfer program cURL soon discontinued its bug bounty program overwhelmed by what creator Daniel Stenberg described as “AI slop.”

“In the old days, somebody put a long time (in) a safety report,” Stenberg said at a recent meeting. “There was a constant debate, but now there is nothing that can be done about this.”

It’s really disappointing because many open source projects are also seeing the benefits of AI tools for writing. Kempf says it’s made creating new VLC modules easy, as long as there’s a professional at the helm.

“You can give the brand the entire VLC codebase and say, ‘I’m porting this to a new machine,'” Kempf said. “It’s good for senior people to write new code, but it’s hard to control people who don’t know what they’re doing.”

Competition ahead

The biggest problem with open source projects is the difference in priorities. Companies like Meta value new code and products, while open source software focuses on stability.

“The problem is different from large companies to open source,” Kempf said. “They get promoted for writing code, not maintaining it.”

AI writing tools are also arriving at a time when software, in particular, is fragmented.

The founder of the Open Source Index, Konstantin Vinogradov, who recently launched an open source initiative, said that AI tools are moving in the same old ways of open source systems.

“On the one hand, we have codes that are growing more and more with increasing interdependence, and on the other hand, we have more maintainers, which may be growing slowly, but not yet,” said Vinogradov. “With AI, both sides of this equation have grown.”

It’s a new way of thinking about AI’s impact on software engineering — one that has implications for the entire industry.

If you see engineering as a way to create functional software, AI scripts make it easier than ever. But if engineering is the process of managing software problems, AI coding tools can make it harder. Little by little, it will take a lot of planning and work to ease the mounting pressure.

For Vinogradov, the result is a familiar phenomenon of open source projects: too much work to do, and not enough good engineers to do it.

“AI is not increasing the number of skilled, skilled caregivers,” he said. “It gives good energy, but all the important problems are still there.”