Agent-Native Products and the Infrastructure Behind AI Autonomy

Three separate signals today point at the same underlying tension: the tooling for running AI agents autonomously is maturing fast (AWS AgentCore, Andrej Karpathy's product design thread), but the people building and training those agents are already grappling with what "rarer human input" actually means in practice — and Amanda Askell's comments at Bloomberg make that concrete in a way that's worth sitting with.