Yumin in the Loop

Yumin in the Loop https://www.yuminloop.com Deep dives into agent frameworks, Microsoft AI patterns, and practical engineering for intelligent systems. Fri, 02 Oct 2026 00:00:00 GMT en Decision models: the small judgments inside useful AI workflows https://www.yuminloop.com/blog/decision-models https://www.yuminloop.com/blog/decision-models Why decision models converged so quickly across Jev, OpenAI, Perplexity, and Cloudflare, and where bounded judgments help real workflows. Thu, 01 Oct 2026 00:00:00 GMT Yumin Zhao decision-models agents architecture evaluation The Road to Autopilots Is a Stack, Not a Ladder https://www.yuminloop.com/blog/autopilots-stack-not-ladder https://www.yuminloop.com/blog/autopilots-stack-not-ladder A four-layer architecture model that separates chat, API agents, harness runtimes, and Microsoft Autopilots by who owns each responsibility. Thu, 03 Sep 2026 00:00:00 GMT Yumin Zhao agent-architecture agent-runtimes microsoft-foundry autopilots identity Running local models inside Microsoft's agent sandboxes https://www.yuminloop.com/blog/local-models-in-agent-sandboxes https://www.yuminloop.com/blog/local-models-in-agent-sandboxes A small ONNX model ran locally in two Microsoft agent runtimes. The ceiling on what you can put there is packaging, not compute. Thu, 20 Aug 2026 00:00:00 GMT Yumin Zhao onnx agent-runtime small-language-models copilot-studio microsoft-365-copilot Four agent runtimes, measured from the inside https://www.yuminloop.com/blog/agent-runtimes-measured https://www.yuminloop.com/blog/agent-runtimes-measured Claude Cowork, Microsoft 365 Copilot Cowork, the GitHub Copilot harness, and Codex run tasks on very different machines. Here is what each runtime actually provides. Tue, 18 Aug 2026 00:00:00 GMT Yumin Zhao agents agent-runtime sandbox copilot-studio microsoft-365-copilot anthropic architecture