Summary
Microsoft’s Ignite event introduced seven new models, but this analysis argues the most important announcement was not a model at all. Sam Witteveen looks at how Microsoft is betting on the harness, meaning the agent loop, tools, context, routing, and sandbox around a model, rather than on any single model.
The best coding result of the week came from Copilot Fusion, a system that picks and coordinates other models. Fusion Max scored 71.2% on Terminal Bench 4.1, more than five points ahead of Claude Opus 5.5 and Fable 5.2 at a fraction of the cost. Microsoft also claimed 96% of Opus 5.5’s performance at 5% of the cost, though the video notes those claims did not come from the newly announced models.
The video covers Satya Nadella’s description of Copilot as the new operating system for work, the Foundry optimizer that improved an airline demo agent by rewriting prompts and tool descriptions, Microsoft IQ for context engineering, and a partnership with Nous Research. It also examines where Microsoft’s own MAI models fit, mostly in cheaper tiers, and points out a Google model that was showcased without being credited on stage.
📺 Source: Sam Witteveen · Published November 20, 2026
🏷️ Format: News Analysis







