Fetching from the wire…
01
02
03
04
05
06
07
08
Public story · 2026-08-04 · high
SWE-Touch injects conflicting edits mid-task, and models fail by keeping or blindly overwriting the conflict.
Why now: This finding is part of the August 4, 2026 coverage, and it's the spread across nine models and three benchmarks that makes it hard to write off as a one-off.
Coding agents lose 7.7 points of resolve rate when a human edits their target files mid-task, per a new benchmark called SWE-Touch. I edit files while Claude Code works in the background on almost every project I ship, so this isn't abstract.
SWE-Touch mines task-critical code regions from past repair trajectories. It builds a plausible Counter-Edit that conflicts with the task. Then it drops that edit into the repo with a contextual message right as the agent reaches that region.
The drop holds across nine models on SWE-bench Verified. It isn't one model's quirk, since the same failure pattern persists on SWE-Bench Pro and DeepSWE, detailed in arXiv:2608.02499.
Trajectory analysis found two failure patterns. Some agents keep the conflicting code exactly as it landed. Others overwrite it and move on without re-checking the rest of the repo.
The paper doesn't test whether forcing a fresh repo diff before every write would close the gap. But the failure pattern points at a process problem. Agents that never re-inspect the repo after a conflicting edit lands are the ones losing points, regardless of which model runs the loop.
Each link below shares sources, entities, or timing with this story.
Shared entities / Same source domain / Shared topic / Earlier coverage
Both cover DeepSWE, SWE, Verified; reported by the same outlet (arxiv.org); overlapping topics (agent, coding, swe bench).
Both cover Coding, SWE, Verified; reported by the same outlet (arxiv.org); overlapping topics (agent, coding).
Both cover SWE, Verified; reported by the same outlet (arxiv.org); overlapping topics (agent, average, code, coding).
Shared entities / Shared topic / Earlier coverage / Tension
Both cover Bench Pro, SWE, Verified; overlapping topics (agent, coding); earlier Bench Pro coverage from 2026-02-12.
Shared entities / Shared topic / Earlier coverage
Both cover DeepSWE, SWE, Verified; overlapping topics (agent, coding, point); earlier DeepSWE coverage from 2026-05-27.
Shared entities / Same source domain / Shared topic / Tension
Both cover Bench Pro, SWE; reported by the same outlet (arxiv.org); overlapping topics (code, coding, completion).
Shared entities / Earlier coverage
Both cover Bench Pro, DeepSWE, SWE, Verified; earlier Bench Pro coverage from 2026-07-21.
Shared entities / Shared topic / Earlier coverage
Both cover Bench Pro, SWE, Verified; overlapping topics (agent, code); earlier Bench Pro coverage from 2026-03-06.