AI cuts the time it takes to write code by 52%, but a quarter of pull requests are still described as low-quality "AI slop", according to a new Linux Foundation report.
A pull request is a proposed code change submitted to a software project. Maintainers, the people who review and merge those changes, still have to check each one for quality and security. The researchers found that more code now means more human reviews, not fewer.
The report is not all bad news. 71% of maintainers agree AI has lowered the barrier to entry for contributing. Some easier review tasks are starting to benefit from the technology too. But 21% of respondents said AI increases the time needed for more complex review work. AI-generated bug reports have also produced false positives and duplicate submissions, which eat reviewer time without improving the software.
The handoff between writer and reviewer looks shaky. Only 61% of respondents said they conduct detailed code reviews of their own work before submitting it. Not every developer discloses AI use to the next person in the chain, so a reviewer often cannot tell how closely to look.
The pattern is a familiar one: speeding up one stage of a process moves the bottleneck rather than removing it. Writing code got cheaper. Judging whether it is any good did not, and that job falls on a small number of people, often volunteers. The report points to a shift from developer burnout to maintainer burnout.
The Linux Foundation is not asking projects to back away from AI. Hilary Carter, its Research SVP, called for "setting clear expectations for GenAI, putting responsibility for testing and disclosure on contributors and giving maintainers better tools to manage incoming pull requests." Mark Collier, its AI & Infrastructure GM, added that "the answer is not to slow innovation, but to build the governance, transparency and trust mechanisms" for AI-assisted contributions.
The open question is who builds those tools. Disclosure rules only help if contributors follow them, and the 61% figure suggests that basic self-review is not yet universal.