Category: AI-Ops

September 20, 2026 · Weekly Edition AI Ops Three stories this week carry production numbers somebody measured. Uber stripped 50,000–70,000 tokens of MCP tool schema out of every agent session and took cost per 1,000 agent requests down about 34%. DoorDash turned multi-agent LLMs loose on 60,000 feature flags and…

September 13, 2026 · Weekly Edition AI Ops A thin week for evidence and a heavy one for marketing. Five items in this issue carry numbers anyone measured: AWS cut P50 time-to-first-token 71–77% with prefix-aware routing on SageMaker (its own benchmark, on its own hardware); Netflix rebuilt Conductor to 420…

September 6, 2026 · Weekly Edition AI Ops A week of measurements rather than announcements. AWS took a GPU cold start apart phase by phase and got eight minutes down to under thirty seconds on a warm node; Google Cloud benchmarked Gemma 3 on TPU v6e and found a hard…