Last week, I ran an experiment that failed. The hypothesis was simple: syllogistic prompts ("Major...
My Experiment Showed Zero Effect. A Statistician Told Me My Measurement Was Broken.
Related
Notable this week: Kimi K3 weights land, MCP goes stateless, OfficeCLI for agents
Five releases from July 28–August 1: Kimi K3 open weights on HuggingFace, MCP 2026-07-28 stateless spec, OfficeCLI for agents, GLM-5.2 comparison, and VS Code Copilot updates.
Five things I noticed this week: GPT-5.6, Gemini Robotics 2, and GitHub stacked PRs
Recap for July 28–31: GPT-5.6's cost-efficiency push, Gemini Robotics 2, GitHub stacked PRs preview, the 2x-not-10x LLM debate, and Kedge's SQLite-first infra bet.
Make agent-callable writes idempotent, or lose data
The unglamorous contract — idempotency keys, server-side dedup, safe retries, and typed errors — that decides whether an agent-native product survives production. With TypeScript p...