The most useful mental model we’ve found for choosing an AI isn’t technical at all. It’s the one you already use at work: different colleagues are good at different things, and you don’t take a design question to the accountant. Each model was trained by different people, on different data, with different tastes — and it shows, exactly the way upbringing shows in people.
You don’t ask “which colleague is best?” You ask “who do I go to for this?”
Claude — the straight-talking designer
Go to Claude when you want something made. It’s the colleague who sketches while you’re still describing the problem: strongest at visual work, UI, design taste and code that looks the way you hoped, and it talks to you straight rather than telling you what you want to hear. Ask it for “something I can look at” and you’ll usually get something surprisingly close to finished.
ChatGPT — the thinking partner
Go to ChatGPT when you want company in the problem. It’s the colleague who pulls a chair over: happy to explore an idea for an hour, crunch numbers, work through analytics, plan a project alongside you and adapt as you change your mind. Where Claude hands you a thing, ChatGPT works the problem with you — a partner more than a producer.
Grok — the one who checks the room
Go to Grok when you want the collective view, right now. Wired into what people are currently saying, it’s more focused on the crowd than on you — which means it’s unusually willing to tell you you’re wrong when other people disagree. Useful for what’s happening today, and for the specific blunt answer rather than the diplomatic one.
The cheat sheet
- “Make me something visual, quickly, and be straight with me” — Claude.
- “Work through this with me — ideas, numbers, plans” — ChatGPT.
- “What’s happening right now, and am I wrong about this?” — Grok.
Why this matters
Because most disappointment with AI is a casting error. People ask one model for everything, get a mediocre answer to the thing it’s worst at, and conclude the technology is overhyped. The fix costs nothing: keep two or three of these colleagues around, learn their personalities the way you learned your teammates’, and route the question to the right desk. The models will change — they get reshuffled every few months — but the habit of casting them well is permanent.
And if you’re wondering how any of these colleagues actually think — they’re all playing the same game underneath. We built a little exhibit that shows it, one word at a time, on our front page.