A world that runs on increasingly powerful AI coding tools is one where software creation is cheap, or so the thinking goes. This idea leaves little room for traditional software companies. As one analyst report put it, vibe coding will allow startups to replicate the features of complex SaaS platforms. This has led to hand-wringing and declarations that software companies are doomed.
Open source software projects that use agents to address long-standing resource constraints should logically be among the first to benefit from this era of cheap code. But that equation just does not quite hold. In practice, the impact of AI coding tools on open source software has been far more mixed.
Industry experts report that AI coding tools have caused as many problems as they have solved. The easy-to-use and accessible nature of these tools has enabled a flood of bad code that threatens to overwhelm projects. Building new features is easier than ever, but maintaining them is just as hard. This trend threatens to further fragment software ecosystems.
The result is a more complicated story than simple software abundance. Perhaps the predicted, imminent death of the software engineer in this new AI era is premature.
Across the board, projects with open codebases are noticing a decline in the average quality of submissions. This is likely a result of AI tools lowering barriers to entry. Jean-Baptiste Kempf, the CEO of the VideoLAN Organization that oversees VLC, said in a recent interview that for people new to the VLC codebase, the quality of the merge requests is abysmal.
Kempf remains optimistic about AI coding tools overall but says they are best for experienced developers. There have been similar problems for Blender, a 3D modeling tool maintained as open source since 2002. Blender Foundation CEO Francesco Siddi said LLM-assisted contributions typically wasted reviewers’ time and affected their motivation. Blender is still developing an official policy for AI coding tools, but Siddi said they are neither mandated nor recommended for contributors or core developers.
The flood 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 vouched users, effectively closing the open-door policy for open source software. As Hashimoto put it, AI eliminated the natural barrier to entry that let open source projects trust by default.
The same effect has emerged in bug bounty programs, which give outside researchers an open door to report security vulnerabilities. The open source data transfer program cURL recently halted its bug bounty program after being overwhelmed by what creator Daniel Stenberg described as AI slop. Stenberg said that in the past, someone actually invested a lot of time in a security report. There was a built-in friction, but now there is no effort at all in doing this. The floodgates are open.
It is particularly 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 far easier, provided there is an experienced developer at the helm. He noted that you can give the model the whole codebase of VLC and ask it to port to a new operating system. It is useful for senior people to write new code, but it is difficult to manage for people who do not know what they are doing.
The bigger problem for open source projects is a difference in priorities. Companies like Meta value new code and products, while open source software work focuses more on stability. Kempf commented that the problem is different from large companies to open source projects. Developers at large companies get promoted for writing code, not maintaining it.
AI coding tools are also arriving at a moment when software, in general, is particularly fragmented. Open source investor Konstantin Vinogradov says AI tools are running into a long-standing trend in open source engineering. On one hand, we have exponentially growing code bases with exponentially growing numbers of interdependencies. On the other hand, we have a number of active maintainers which is maybe slowly growing, but definitely not keeping up. With AI, both parts of this equation accelerated.
This is a new way of thinking about AI’s impact on software engineering, one with alarming implications for the industry at large. If you see engineering as the process of producing working software, AI coding makes it easier than ever. But if engineering is really the process of managing software complexity, AI coding tools could make it harder. At the very least, it will take a lot of active planning and work to keep the sprawling complexity in check.
For Vinogradov, the result is a familiar situation for open source projects: a lot of work to do, and not enough good engineers to do it. He remarked that AI does not increase the number of active, skilled maintainers. It empowers the good ones, but all the fundamental problems just remain.