Summary
Chris Raroque attends Microsoft Build in San Francisco — the first time the conference was held there — and documents what changed his mind about local AI models running directly on consumer Windows hardware. Arriving as a skeptic building primarily for the Apple ecosystem, he leaves convinced that Windows is meaningfully ahead on on-device AI for practical developer use cases.
The clearest demonstration of that shift is a real-time video upscaling demo running entirely on a standard Windows laptop with no server involvement — a capability Raroque notes would normally require expensive cloud-based upscaling infrastructure. He also encounters Microsoft Aurora, a foundation model designed not for text generation but for forecasting Earth systems including weather and air pollution, expanding his mental model of what foundation models can do. The conversation with the Microsoft Windows AI team (specifically a developer named Greg) produces the most technically grounded segment of the video.
A conversation with Paul, an engineer on the Logitech SDK team, surfaces a broader ecosystem signal: because of tools like Claude Code, the audience for hardware SDKs is fundamentally changing. Logitech now targets individual builders rather than large enterprise partners, and has published an LLMs.txt file so coding agents can learn the SDK directly. Raroque also touches on Google’s Gemma 4 as an open model worth tracking, and reflects on how Apple’s on-device foundation models still lag behind Windows in developer-facing capability for his use cases.
📺 Source: Chris Raroque · Published June 22, 2026
🏷️ Format: Showcase
