Which AI to use: Free users don’t get the good AI, and even if you pay, you have to pick thinking mode. Once you’ve done that, the model matters less than the harness. For newsrooms: curated archive retrieval, a system prompt encoding editorial voice, CMS integration, verification. (Ethan Mollick, One Useful Thing)
- Mollick's eighth AI guide since ChatGPT introduces a new framework: models (the brains), apps (the interfaces), and harnesses (the systems that let AI actually do things). The same Claude Opus 4.6 behaves very differently in a chat window than inside Claude Code running autonomously for hours.
- The free models are optimized for speed and personality, not accuracy. For real work, pay the $20 and manually select the advanced model, because "auto" mode quietly downgrades you.
- The shift from chatbot to agent is the biggest change since ChatGPT launched. An AI that does things is more useful than one that talks about them, and most organizations are still treating this as a tooling decision when it's already an operating model question.
