Sam Schillace

Corporate Vice President, Deputy CTO - Consumer Experiences and Productivity — Microsoft Engineer, Google Docs creator — Google Founder — Writely

Listen — profile

0:00 / 3:05

Profile

Sam Schillace occupies a rare vantage point in software: he built one of the defining collaborative tools of the Web 2.0 era, and two decades later he is helping steer one of the largest companies in tech through the LLM transition. He is Corporate Vice President and Deputy CTO at Microsoft, part of CTO Kevin Scott’s organization, where he focuses on consumer products, productivity, and the developer tooling a world of AI-native software will need. If you want to understand how the people who lived through the last platform shift think about this one, Schillace is one of the most useful voices to follow.

His origin story is the one developers know him for: in the mid-2000s he co-founded Writely, a browser-based word processor that Google acquired in 2006 and turned into Google Docs. At the time, the idea that a real document editor could live in a browser and let multiple people type into the same file simultaneously was close to heresy — and it reshaped how a generation collaborates. He went on to run large swaths of Google’s consumer portfolio (Gmail, Blogger, Picasa, Reader, parts of Maps), invested at Google Ventures, and served as SVP of Engineering at Box through its IPO. By his own count he is a six-time founder. That is a career of shipping, not theorizing.

Since joining Microsoft in 2021, Schillace’s most influential contribution to the AI community has been intellectual rather than a single product. With early access to GPT-4 in 2022, he began the incubation that became Semantic Kernel, Microsoft’s open-source SDK for wiring LLMs into applications — an early counterpart to LangChain. Out of that hands-on work came the “Schillace Laws of Semantic AI,” a set of nine hard-won heuristics for building with models: don’t write code if the model can do it; the model will get better but your code won’t; code is for syntax and process, models are for semantics and intent; text is the universal wire protocol; ask smart to get smart. These circulated widely (Simon Willison among others amplified them) because they distilled, unusually early, the mental shift required to program with probabilistic systems instead of deterministic ones.

For a developer learning AI, Schillace matters because he treats LLMs as a genuinely new engineering substrate rather than a chatbot novelty — and he’s blunt about the messiness of that transition. His ongoing “Sunday Letters” newsletter is where he thinks in public about AI coding, organizational change, taste, and why abundant, cheap code changes what engineers are actually for. He is a self-described “messy tech optimist,” which is worth the honesty: he’s bullish, but he’s lived through enough hype cycles to know that innovation is nonlinear and mostly failure.