Why Agentic-PRs Get Rejected: A Comparative Study of Coding Agents CCS: Software and its engineering Software repository mining

Sota Nakashima, Yuta Ishimoto, Masanari Kondo, Shane Mclntosh, Yasutaka Kamei · 2026

Read Why Agentic-PRs Get Rejected: A Comparative Study of Coding Agents CCS: Software and its engineering Software repository mining by Sota Nakashima, Yuta Ishimoto, Masanari Kondo, Shane Mclntosh, Yasutaka Kamei (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2602-04226.txt.

Description

Coding agents submit pull requests that humans reject for agent-specific reasons. Sota Nakashima and 4 co-authors inspected six hundred fifty-four rejected agentic pull requests from five coding agents, plus a human baseline, in a 2026 study. Earlier work covered one agent's rejections only.

Seven rejection modes appeared only in agentic submissions, including outright distrust of AI-generated code. Patterns differed by agent, such as one agent automatically withdrawing its own inactive requests. Agents are chosen for different purposes, so their failure fingerprints differ. Maintainers and agent builders alike get a map of where trust in automated code breaks down.

Catalog record

CREATORSota Nakashima, Yuta Ishimoto, Masanari Kondo, Shane Mclntosh, Yasutaka Kamei
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE22 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2602-04226

Browse the full library · About the Tower of Babel