SWE-Shift finds coding agents stumble when code changes underneath them
What happened
Computing Life reported on Oct 29 that git worktree isolates an AI agent's working directory but does not remove the risk of interface changes at merge time: an agent following an old plan can still delete new logic on the main branch. The article cites SWE-Shift tests across 17 coding systems and 120 tasks, showing that passing existing tests does not guarantee new code survives. Clearing old conversation and rescanning the workspace raised paired success by 6 to 9 percentage points.
Written by AI from the coverage · updated 1 hour ago
Coverage
Follow the reports to see the story from different sides.
- Computing Life · Share · YagePickWhy git worktree isolation doesn't stop AI agents from deleting new logic at merge time
git worktree isolates an AI agent's working directory, but interface changes at merge time can still lead an agent following an old plan to delete new logic on the main branch. The article cites SWE-Shift tests across 17 coding systems and 120 tasks: passing existing tests does not guarantee new code survives, and clearing old conversation plus rescanning the workspace raised pairwise success by six to nine percentage points.
Heat over time
Not enough continuous observations to draw a trend yet.