Coding agents spend 60.5% of their documentation time on instruction files and working notes.

Shared entities / Same source domain / Shared topic / Earlier coverage

Coding Benchmarks Conflate Model, Retrieval, and Harness

Both cover Coding, SWE; reported by the same outlet (arxiv.org); overlapping topics (agent, agentic, coding).

Shared entities / Shared topic / Earlier coverage

Berkeley Researchers Score 100% on SWE-bench Verified Without Solving a Single Task. Every Major AI Benchmark Is Broken.

Both cover AGENT, Everyone, SWE; overlapping topics (agent, coding); earlier AGENT coverage from 2026-04-12.

Shared entities / Same source domain / Shared topic / Earlier coverage

Coding agents spend 631K tokens hunting for the right file

Both cover Coding, SWE; reported by the same outlet (arxiv.org); overlapping topics (agent, coding).

Coding agents lose 7.7 points when you edit files mid-task

Both cover Coding, SWE; reported by the same outlet (arxiv.org); overlapping topics (agent, coding).

PerfAgent triples optimization success by reading profiler traces

Both cover Coding, SWE; reported by the same outlet (arxiv.org); overlapping topics (agent, coding).

SWE-bench Verified is leaking answers, and 60.8% of the "solved" issues prove it

Both cover Coding, SWE; reported by the same outlet (arxiv.org); overlapping topics (agent, coding).

Your CLAUDE.md is doing something. Whether it helps depends entirely on how you wrote it.

Both cover AIDev, Everyone; reported by the same outlet (arxiv.org); overlapping topics (agent, agentic).

AI Coding Agents Don't Write Clean Patches. They Can't.

Both cover Agent, SWE; reported by the same outlet (arxiv.org); overlapping topics (agent, coding).