No Priors Ep. 18 | With Kevin Scott, CTO of Microsoft

Scott stresses that simply “adding an LLM” is not enough; the most important products will be those that turn previously impossible tasks into hard but feasible ones, in the same way smartphones enabled non-obvious apps like TikTok and DoorDash rather than just early novelty apps.

Microsoft stopped “peanut butter spreading” GPUs and instead centralized capital-intensive compute around high-conviction AI efforts, enabling the scale necessary for frontier models like GPT‑3 and beyond.

From GitHub Copilot to Microsoft 365 and Bing Chat, Scott describes a generalized copilot pattern: LLMs orchestrated with tools, retrieval, prompts, plugins, and safety filters to assist humans in domain-specific workflows rather than replace them.

Real-world deployments already use multiple models for cost, latency, and quality tradeoffs; Scott is excited by open-source innovation but notes that robust safety and responsible AI practices must evolve alongside it.

Tasks that once required deep ML expertise and months of work can now be done in hours by far less specialized users; Scott sees this accessibility as a path to more equitable opportunity for people far from traditional tech hubs.