Reading
Curated notes on things worth reading — source, link, and a short take.
-
Inside vLLM: why serving shape matters
Paged KV-cache and continuous batching are the real product — not a list of model names.
-
Stateful benchmarks for long-horizon agents
Scores only mean something when harness, runtime, and task set are pinned.
-
Humans miss agent permission threats
Approve/deny UX is a safety surface — fatigue and familiar wrappers hide exfiltration.