GitHub's Jay Parikh says agents draft 40% of pull requests

In an interview released on November 7, Jay Parikh, executive vice president of Microsoft CoreAI and the head of GitHub, said that the Pull Request should not be a required step for every developer. For personal projects meant to test an idea, he said he does not even create a Pull Request or manage Issues, preferring to turn the idea directly into code and keep building. For large enterprise projects, however, code review and permission management remain essential.

Parikh backed up the shift with GitHub's own numbers. As of September, more than 40% of Pull Requests on the platform had their first draft written by a Coding Agent. Back in March, that figure was only 17%, so the share of agent-drafted first versions rose roughly sixfold in six months. Parikh said the growth reflects a broader change in how code is produced: when more of it is generated by agents, workflows designed around human developers need to be revisited.

The conversation, published on the Latent Space podcast, also covered multi-model collaboration, code review practices, software supply chain security, and the economics of AI-assisted coding. On the idea of using AI to rewrite large amounts of open source dependencies and thereby reduce supply chain risk, Parikh cautioned that even if an agent can rewrite the code for you, the ongoing maintenance work does not disappear.

No platform-wide change has been announced yet, but Parikh's stance is already clear. He no longer uses Pull Requests in every project himself, and he suggested that teams should treat the process as conditional rather than universal, matching review practices to the stage of a project and the maturity of the team.