AI Radar

Noteworthy AI conversations and writing, summarized from publisher notes.

Microsoft's Kevin Scott on the Agentic Web

· AI & I

Microsoft CTO Kevin Scott argues that 2025's defining theme is the "agentic web"—an internet-like ecosystem where agents can take action on users' behalf, calling APIs, consulting data sources, and making changes across systems. He credits this shift to two developments: scaling laws have continued to hold (so they no longer dominate conversation), and a "capability overhang" has emerged where model reasoning ability outpaces what products actually deliver. He positions protocols like MCP as the "HTTP of the agentic web" and NL web as its HTML, and pushes for all Microsoft-internal systems to speak a standard protocol to agents so the company stops "shipping its org chart." On security, Scott acknowledges MCP's model is still nascent and calls for agent identities, entitlement systems, and user permission flows to be built openly rather than as proprietary moats. He rejects the false dichotomy between open permissionless innovation and walled-garden security, arguing AI-powered personal security agents can provide robust protection in open systems. He also pushes back on the idea that coding agents erode software craftsmanship, drawing an analogy to debates over power tools and CNC machines among woodworkers—makers will always have strong opinions about their tools, but curiosity matters more than orthodoxy. Scott predicts a future with many competing, differentiated agents rather than a single dominant one, and within a year expects a shift from synchronous to asynchronous agent interactions, where users delegate tasks and return to completed work rather than prompting and waiting. He warns that anyone treating current cost or capability limits as reasons to delay adoption will fall behind as both continue improving.

AI Radar wrote this summary from the publisher’s notes with AI, not from a transcript. How it works.