← The Polylog AI Intelligence Brief
Morning Edition · Friday, September 11, 2026Published at 2:22 AM EDT · New York
Researchers Find Agent Skills Work Better as Subagents Than as Context Injections
A Cornell and Microsoft Research paper argues fresh context windows beat loading skill instructions into a growing main context, when skills expose clear input-output contracts.
A team from Cornell University and Microsoft Research, led by Wasu Top Piriyakulkij, examines a question every agent developer encounters in practice: when an agent has a library of reusable skills, should it read the skill instructions its…
Continue the AI Intelligence Brief
Track frontier labs, chips, export controls, model releases, regulation, and AI infrastructure.
- 5 AI intelligence signals a day
- Frontier labs, compute, and chips
- Model releases and AI infrastructure
- Source-grounded analysis with confidence labels
The Global Intelligence Brief stays free.
Part of a tracked trend
Agent Context Architecture Standardizes
Reliability on long-horizon agent tasks is increasingly determined by how context is partitioned rather than by model capability, so expect recurring architectural conventions, contracts and registry requirements to form around context isolation.
Townsquare
Start a discussion in Townsquare.
More from this edition
- Anthropic Ties 151 Million Claude Exchanges to an Alleged Alibaba Distillation Campaign
- OpenAI Opens the Codex Harness as a Managed Agents API
- OpenAI Pauses New ChatGPT Pro Sign-Ups Because Astra Demand Exceeds Its Compute
- Claude Fable 5.1 Posts Large Gains on Terminal and Automation Benchmarks
- OpenAI and the General Services Administration Cut Federal ChatGPT License Fees to Zero
- OpenAI Aims GPT-6 Astra at Investment Banking Workflows
- d-Matrix Will Plug Its Inference Chips Into NVIDIA Racks Through NVLink Fusion
- Skild AI's S1 Learns Robot Tasks From One Video Without Weight Updates
Comments
0No comments yet.