tl;dr: Ethan Mollick got early access to Claude 5 Fable, Anthropic's first "Mythos-class" model, and found it represents a major leap—autonomously spawning sub-agents to research, code, and verify work over multi-hour sessions, producing things like a fully-researched isochrone map and a 9.5-hour build of sophisticated research software. The downside is cost (twice Opus, heavy token burn), aggressive security guardrails, and a deeper shift in the human role: users no longer steer the process but commission outcomes from an opaque black box.