Mining agent playbooks from telemetry instead of hand-authoring them.

Shared entity: Microsoft Research / Same source domain / Shared topic / Earlier coverage

Microsoft SkillOpt Trains Markdown Skills for Frozen LLMs. 52/52 Benchmark Wins. Your CLAUDE.md Can Be Optimized Like a Neural Network.

Both cover Microsoft Research; reported by the same outlet (arxiv.org); overlapping topics (agent, builder).

Shared entity: Microsoft Research / Same source domain / Earlier coverage / Tension

RefineRL: 4B Models Match 235B via Self-Refinement RL.

Both cover Microsoft Research; reported by the same outlet (arxiv.org); earlier Microsoft Research coverage from 2026-04-21.

Shared entity: Microsoft Research / Same source domain / Earlier coverage

Microsoft Research: LLMs Silently Corrupt 25% of Your Documents in Long Workflows

Both cover Microsoft Research; reported by the same outlet (arxiv.org); earlier Microsoft Research coverage from 2026-05-10.

CUWM: World Models for Desktop Agents (arXiv 2602.17365)

Both cover Microsoft Research; reported by the same outlet (arxiv.org); earlier Microsoft Research coverage from 2026-02-22.

Same source domain / Shared topic / Tension

Your AI-Generated Code Compiles. It Doesn't Work.

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

PostTrainBench: Agents Autonomously Post-Train LLMs — But Reward Hack When Unsupervised

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

Security Threat Modeling: MCP vs A2A vs Agora vs ANP

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

Shared entity: Microsoft Research / Earlier coverage / Tension

Uber Burned Its Entire 2026 Claude Code Budget by April. The COO Says He Can't Justify the Spend.

Both cover Microsoft Research; earlier Microsoft Research coverage from 2026-05-26; pushes against this story (but).