ORAgentBench: the best agent solves 35.51% of operations-research tasks, 20.59% of hard ones.

Shared entity: Failures / Same source domain / Shared topic / What happened next / Tension

Best models hit 70.4% on single-hop API calls and 2.4% on knowing a question is unanswerable.

Both cover Failures; reported by the same outlet (arxiv.org); overlapping topics (agent, benchmark, best).

Shared entity: Failures / Same source domain / Shared topic / What happened next

Forced revision makes coding agents worse. 0.820 → 0.673.

Both cover Failures; reported by the same outlet (arxiv.org); overlapping topics (agent, code).

Same source domain / Shared topic / Tension

RSIBench-Data: agents improve on their first attempt 58% of the time but finish below their own peak in 78% of runs.

Reported by the same outlet (arxiv.org); overlapping topics (agent, benchmark, best, code); pushes against this story (but).

The best frontier agent clears 15% of real terminal tasks. Read that again.

Reported by the same outlet (arxiv.org); overlapping topics (agent, autonomou, best, task); pushes against this story (against).

Same source domain / Shared topic / Tension / Downstream implication

Test your RAG under corpus mutation, not against a static snapshot.

Reported by the same outlet (arxiv.org); overlapping topics (best, check); pushes against this story (against).

Same source domain / Shared topic / Tension

AI agents followed written policy just 36.2% of the time

Reported by the same outlet (arxiv.org); overlapping topics (agent, autonomou, check); pushes against this story (against).

SWE-NFI grades AI refactors on structure, not tests alone

Reported by the same outlet (arxiv.org); overlapping topics (agent, best, code); pushes against this story (against).

Run a dedicated spec-elicitation agent before the coding agent.

Reported by the same outlet (arxiv.org); overlapping topics (agent, benchmark, code); pushes against this story (against).