An agentic pull request (Agentic-PR) is a pull request submitted to a repository by an AI coding agent, so that the agent participates directly in real-world software development. Why Are Agentic Pull Requests Merged or Rejected? An Empirical Study uses the term for agent-authored PRs to open-source repositories, drawing on evidence from the AIDev dataset that such PRs span thousands of repositories, and treats a closed Agentic-PR as one that received an explicit merge or rejection decision.
Usage
The term is used in mining-software-repositories research as the unit of analysis for studying how agent-authored changes fare once they meet human reviewers. In that study’s account, Agentic-PRs are commonly evaluated with outcome-based measures such as merge rates or approval frequencies, and the study’s central claim is that those outcomes misrepresent agent capability on their own: a rejection may reflect a workflow constraint or an undocumented reviewer decision rather than an agent failure, and a merge may depend on reviewer feedback or commits applied by a human. In its manually inspected sample only 35.7% of rejected Agentic-PRs showed clear agentic failures, and 15.4% of merged ones involved explicit reviewer participation.
Because of this, the study separates Agentic-PRs reviewed only by bots from those with human comments, excluding the former as cases where no human decision context is available, and argues for interaction-aware evaluation that reads reviewer comments, CI outcomes, commit history and workflow actions rather than the final PR state alone.