WorkThe story, in brief
AI coding agents find the right file but miss the exact lines that matter, study shows
Claude Code finds the file. Misses the fix. New SWE-Explore benchmark exposes the gap that's silently killing AI agent deployments.
The KeyNews take
Why it matters
A new benchmark reveals a critical blind spot in production AI coding agents: they excel at file-level search but fail at line-level precision. This matters for leaders betting on autonomous code repair—the capability gap is wider than marketing claims suggest.
The key facts
4 to knowSWE-Explore is first benchmark to separately test code search from code repair
AI agents (Claude Code, Codex tested) reliably locate correct files but miss critical lines within files
Context window limitations identified as root cause of repair failures
Benchmark finding suggests silent failure mode in production deployments
Go to the source
The Decoderthe-decoder.com
Publisher excerpt: AI coding agents like Claude Code or Codex reliably find the right file but miss most of the critical lines within it. The new SWE-Explore benchmark is the first to test code search separately from the actual repair, and it shows that without enough context, even the best fix will fail.