Is This Pull Request Ready for Review? An Empirical Study of Autonomous Coding Agents Pull Requests
Human developers rely on workflow cues to decide when a pull request (PR) is ready to be reviewed. One such cue is the Ready for Review (RFR) status, which is commonly interpreted as indicating that development has reached a state that is ready for review. As autonomous coding agents increasingly author pull requests, examining how this signal appears and is used in agent-authored workflows helps characterize how these agents integrate into modern development processes and the extent to which they operate independently. In this study, we empirically examine the occurrence and properties of RFR events in agent-authored PRs. We analyze 33,596 PRs authored by five autonomous coding agents from the AIDev dataset. By examining transitions to RFR, we study when the signal is applied, who initiates it, and how much development activity precedes it, as well as how PRs with and without RFR differ in their subsequent evolution. Our results show that the use of RFR in agent-authored pull requests remains ad hoc and largely driven by human judgment: only 9.8% of agent-authored PRs transition to RFR, and 94.7% of these transitions are initiated by humans rather than agents. We further observe substantial variation across agents in the amount of development effort required to achieve readiness, ranging from near-instantaneous PR completion to extended pre-review phases with sustained activity. Finally, PRs marked as RFR receive more review interaction and remain open longer than PRs without this signal, reflecting more involved review and integration processes.