The New Bottleneck: Human Review, Not Machine Code
AI coding agents in open source are tools that help developers generate, modify, and submit code changes and pull requests at higher volume, shifting the constraint on project progress from code production to human review capacity and maintainer workload on GitHub, rather than replacing contributors or reducing newcomer participation. The core story is blunt: AI coding agents are not killing open source, but they are flooding it. A new study tracking 1,888 repositories that added configuration files for tools like Cursor and Claude Code found that newcomer participation stayed steady or inched up. The dreaded “crowding out” of beginners is not materializing. What is changing is the number and complexity of pull requests. Merged pull requests across platforms have jumped from about 25 million per month in early 2023 to roughly 90 million a month, a near-quadrupling that lands squarely on core maintainers.
Studies Show More Code, More Complexity—and Still Plenty of Newcomers
The Peking University team treated AI adoption as the moment a repository committed its first agent configuration file and compared those projects to similar ones that never adopted AI coding agents open source tools. Using difference-in-differences—the gold standard for teasing apart cause and correlation—they found that newcomer participation held or rose slightly. Even under conservative assumptions, the worst observed decline was a 1.5% dip that was not statistically significant. Meanwhile, code got more complex. Cyclomatic complexity increased 3% to 4% across languages, and cognitive complexity jumped about 11% in Python projects after AI agents entered the workflow. These higher complexity scores did not scare newcomers away in the 128 Python repositories where complexity clearly rose: newcomer entry did not fall, retention held, and the active contributor base grew. In other words, open source sustainability is not yet threatened by AI chasing humans out—it is threatened by what happens after those humans click “Create pull request.”
GitHub’s Eternal September and Maintainer Workload
If AI coding agents aren’t shrinking the contributor pool, they are definitely swelling the contribution stream—and that is where maintainer workload GitHub pressure spikes. Outbound collaboration, the total pushes and pull requests crossing economic borders, grew by 16% from Q4 2025 to Q1 2026, the second-fastest rise since 2020. That growth shows up as more pull requests, more issues, and more comment threads for maintainers to triage. Contribution volume has strained several communities enough that GitHub’s open source programs director described the situation as an “Eternal September of Open Source,” where the steady arrival of newcomers forces communities to rethink how they operate. To cope, GitHub shipped pull request automation controls: caps on how many open pull requests a non-writer can hold, repo-level toggles that restrict or disable pull requests and issues, pinned comments to keep key guidance visible, and banners that cut down on noisy “+1” replies. These are firebreaks, not long-term answers. The code keeps flowing; the number of maintainers does not.
AI Coding Agents Lift Pull Requests—and Overload Review Queues
On the enterprise side, the numbers are even starker. A large Microsoft study of command-line AI agents—Anthropic’s Claude Code and Copilot CLI—found a 24.0% increase in merged pull requests per engineer per day over four months, with a likely range of 14.5% to 33.7%, but only when developers used the tools regularly. Giving people access without changing habits did not move the needle. In a separate enterprise dataset of 196,212 pull requests at a mid-sized company, a public “2x mandate” drove per-developer throughput from 21.2 to 44.3 merged pull requests per active developer by April 2026, a 2.09x jump. The cost was review stress. As AI-authored pull requests climbed, the share receiving at least one human review dropped from 89% to 68%, while automated AI review coverage rose from roughly 19% to 84% and workload per reviewer doubled. According to this research, “AI coding agents can raise output, but only when teams have the usage habits, codebases, and review capacity to absorb the extra work.”
Designing Sustainable Open Source in an AI-Saturated Future
The real problem is not AI replacing human contributors; it is maintainers drowning in the flood of AI-generated pull requests. Pull request volume has nearly quadrupled, code is more complex, and many contributors do not fully understand what they are proposing. That burden lands on maintainers whose ranks are not growing at the same pace. For open source sustainability, we have to treat review bandwidth as the scarce resource. Before scaling AI coding agents, teams should watch four things: adoption by team, how often each developer uses the tools, how much legacy code they touch, and whether reviewers can handle the extra pull request volume. GitHub has opened a community discussion for maintainers to share which controls work, where gaps remain, and what would help next. Future studies aim to measure how heavily projects rely on AI tools and better study repositories “born with AI.” If we do not build processes and infrastructure around review and pull request automation, AI will not break open source—it will break the people holding it together.
