The Real Cost of One-Model-for-Everything
Most teams pick a default AI — ChatGPT, Claude, Gemini — and use it for everything. That feels like a decision, but it isn't. It's a postponed decision that compounds into wasted time, inconsistent output quality, and a creeping sense that AI isn't delivering what it promised.
The actual problem isn't the tools. It's the absence of routing logic.
Ethan Mollick's opinionated guide to which AI to use for what does something most AI coverage won't: it maps specific task categories to specific models, based on hands-on use rather than benchmark scores or vendor positioning. For operators who need to get work done rather than win arguments about leaderboards, this is the most practically useful AI selection framework published this year.
What Task-Level Routing Actually Looks Like
Mollick's framing shifts the question from