What's this about?
Ethan Mollick's periodic plain-language guide, updated for the agent era: free chatbots are fine for low-stakes chat; for high-stakes questions (medical, legal) pay for Claude Opus/Fable or GPT-5.6 Sol on High thinking; for real work, the only two easy on-ramps are Claude or ChatGPT at $20/month — because both now give the AI a computer, not just a chat box.
What exactly changed since his last guide?
"Using AI" used to mean conversation; now it means delegation. His demo: the same city-builder prompt that made a toy last year produces a dramatically better app via Codex now. The unit of AI use went from "message" to "hours of work done in one go."
Why is it worth noting?
This is the guide normal people actually get forwarded. Its blunt two-option conclusion (and "the names are confusing" jab) is the clearest picture of the consumer AI market right now.