For open source software, AI programming tools are both a blessing and a curse

A world running on increasingly powerful AI coding tools is one in which software is cheap to create — or so the thinking goes — leaving little room for traditional software companies. As one analyst report put it, “Biological programming will allow startups to replicate the features of complex SaaS platforms.”

Pay attention to statements and statements that software companies are doomed to failure.

It stands to reason that open source software projects that use proxies to overcome long-standing resource constraints would be among the first to benefit from the era of cheap code. But this equation does not quite hold up. In practice, the impact of AI coding tools on open source software has been much more mixed.

AI coding tools have caused several problems that have been resolved, according to industry experts. The easy-to-use and accessible nature of AI coding tools has allowed for a flood of bad code that threatens to overwhelm projects. Creating new features is easier than ever, but maintaining them is no less difficult and threatens to further fragment software ecosystems.

The result is a story that is more complex than the abundance of simple programs. The expected imminent death of the software engineer in the new era of artificial intelligence may be premature.

Quality versus quantity

Across the board, projects with open code bases are seeing a decline in the average quality of submissions, likely as a result of AI tools lowering barriers to entry.

“For people who are new to the VLC database, the quality of the merge requests we see is very poor,” Jean-Baptiste Kempf, CEO of the VideoLan organization that oversees VLC, said in a recent interview.

Kempf remains optimistic about AI coding tools in general, but says they are best for “experienced developers.”

There have been similar problems in Blender, the 3D modeling tool that has been maintained as open source since 2002. Blender Foundation CEO Franceso Cidi said LLM-supported contributions typically “waste reviewers’ time and affect their motivation.” Blender is still developing an official policy for AI coders, but Sidi said it is “neither mandatory nor recommended for core contributors or developers.”

The deluge of merge requests has gotten so bad that open source developers are building new tools to manage it.

Earlier this month, developer Mitchell Hashimoto launched a system that would limit GitHub contributions to “secured” users, effectively closing the open-door policy for open source software. As Hashimoto said in the announcement, “AI has removed the natural barrier to entry that allowed OSS projects to be trusted by default.”

The same effect has been seen in bug bounty programs, which give outside researchers an open door to report vulnerabilities. Open source data porting software cURL recently discontinued its bug bounty program after it was overwhelmed by what creator Daniel Steinberg described as “AI regression.”

“In the old days, a person would invest a lot of time [in] “Security report. There was internal friction, but now there is absolutely no effort to do so. The flood gates are open,” Steinberg said at a recent conference.

It’s especially frustrating because many open source projects are also seeing the benefits of AI coding tools. Kempf says it has made building new modules for VLC much easier, provided there is an experienced developer at the helm.

“You can give the model the entire code base for VLC and say, ‘I’m going to port this to a new operating system,’” Kempf said. “It’s helpful for older people to write new code, but it’s hard to manage for people who don’t know what they’re doing.”

Competing priorities

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

“The problem is different between large companies and open source projects,” Kempf commented. “They are promoted to write code, not maintain it.”

AI coding tools also arrive at a time when software in general is particularly fragmented.

AI tools are on a long-term trend in open source engineering, said Konstantin Vinogradov, founder of the Open Source Index, which recently launched an endowment to preserve open source infrastructure.

“On the one hand, we have a significantly growing code base with an increasing number of interdependencies, and on the other hand, we have a number of active maintainers, which is probably growing slowly, but certainly not keeping pace,” Vinogradov said. “Thanks to AI, both parts of this equation have accelerated.”

It’s a new way of thinking about the impact of AI on software engineering, and one that has worrying implications for the industry as a whole.

If you see engineering as the process of producing working software, AI programming makes it easier than ever. But if engineering is actually the process of managing software complexity, AI coding tools may make it more difficult. At the very least, it will take a lot of active planning and work to keep the sprawling complexity under control.

For Vinogradov, the result is a familiar situation for open source projects: There’s too much work to do, and not enough good engineers to do it.

“Artificial intelligence does not increase the number of active and skilled supervisors,” he noted. “It empowers good people, but all the underlying problems remain.”