AI's Role in Open Source Development

A report by the Linux Foundation highlights the significant impact of AI on open source development, noting an increase in raw production, a lowering of the entry threshold, a greater need for review, and exhaustion of a portion of the people involved.

Survey Findings on Contributions

The study, conducted by Sam Boysel of the Linux Foundation and Anni Lai of Generative AI Commons, with a preface by Brian Proffitt of Red Hat, found that 59% of maintainers have observed an increase in the volume of contributions, with the figure rising to 69% in projects with at least 50 contributors.

Quality Concerns in Contributions

About a quarter of the merge requests received are considered to be of low quality, 31% in large projects. Not surprisingly, 52% of maintainers indicate that their review time has increased, the figure rising to 64% in large projects.

Burnout and Verification Challenges

In these latter projects, burnout has become the primary concern, ahead of the quality of contributions. A new manifestation of the "verification tax." Verification, justly, largely relies on trust and mutual inspection, formal attestations developing with the size of the project: 20% in small ones, 41% in large ones.

AI Usage Among Maintainders

Sign of the times: three-quarters of the maintainers surveyed indicated using AI at least partially to examine the contributions received. The use of AI within projects is however very variable. Only 35% have a written policy allowing AI, 10% prohibit it, 23% plan to adopt one, and a third have yet to make any plans.

Policy and Licensing Focus

Existing policies focus mainly on disclosure and contributor responsibility (76% each), much less on licenses (39%). The numbers reflect a contrasting landscape that we have seen on several occasions. Some projects, including Debian and the Linux kernel itself, have no problems with AI-generated contributions, provided that the person submitting them is able to explain them.

AI in Development Stages

Other interesting data are present in the report. For example, AI is used in all stages of development: new features (83%), debugging and testing (78%), refactoring (71%), or documentation (69%). As expected, the main time savings observed are at 52% on code writing, but maintainers mention savings decreasing with the complexity of the project.

Time Savings and Complexity

On complex tasks, 21% indicate spending more time than before. If maintainers estimate at 71% that AI lowers the entry barrier, it also degrades trust within the community for 53% of them, while only 13% see an improvement.

Support for New Contributors

In the same area, 46% think that the need to support newcomers in a project is higher. The report, officially provided by the foundation, contains eight recommendations applicable to any free project. These include establishing a clear and written policy on AI-assisted contributions and on vulnerability reporting, giving priority to transparency and disclosure requirements, shifting the verification burden to contributors, standardizing the practices for attributing AI-assisted contributions, using AI as a maintenance tool, continuing to invest in human support for new contributors, balancing the accessibility gains brought by generative AI and its cost for community trust, and emphasizing quality governance and trust over technical performance improvements.

Recommendations and Implementation

Some points can indeed raise a few eyebrows. For example, the recommendation to use the "Assisted-by:" format in commits is directly referencing the Linux kernel. However, maintainers of this latter have recently decided not to make it mandatory.

Limitations of the Survey

Linus Torvalds, in particular, has been very clear on the matter, with a certain fervor towards a kernel maintainer who advocated for preserving this obligation. Even though the report was closed last June, it does not take into account this change, which remains prior to the publication of the Linux Foundation.

Survey Methodology Concerns

This summary also shows several weaknesses. In addition to its purely declarative aspect, the profile of the maintainers surveyed is particular, the respondents having been recruited via the foundation's channels. 52% of the respondents thus have 11 years or more of experience as contributors and 75% use AI daily, all uses combined.

Dominant Sectors in the Survey

Furthermore, the dominant sectors of activity are operating systems and infrastructure. We are therefore facing a group of experienced developers, far from occasional contributors, which distances this summary view from the open source landscape.

Implications of the Report

The report nevertheless brings several benefits, notably numerical data on a subject previously treated mainly through a few incidents and positions. It also shows that the workload is concentrated on large projects and maintainer teams, that there is an obvious delay in project governance, or that complex tasks show no gains or even take more time.

Shift in Production Costs

Most importantly, the report drives a bit more the nail into the shift of production costs to verification. However, it also raises a question without answering it: how to preserve trust if AI expands the pool of contributors while weakening the signals that allowed judging a contribution before reading it in detail?

Open Source in a Dilemma

Free software finds itself in an uncomfortable position, sitting between two chairs, between systematic review and filtering at the entry for new contributors.

Recommendation Omission

Finally, the report mentions in its summary at the beginning a recommendation it does not reiterate in its final list: "fighting fire with fire." In other words, maintainers should also use AI as the first link in the verification chain to regain time.