Most people approach Claude like a search engine. Short query, expect good answer. That works for "what's the capital of France." It falls apart for anything real.
The thing is, Claude isn't bad at your task. It just doesn't know what you actually want. And it'll never ask — it'll just make its best guess and return something that's technically an answer to your question but not what you needed.
Here's the same request, two ways:
Version 1:
Write me a product description for my app.
Version 2:
Write a product description for a B2B SaaS app called "Stackwise"
that helps engineering managers see which internal tools their team
actually uses, and cut the ones they don't.
Audience: CTOs and VPs of Eng at 50–200 person startups.
Tone: Direct. No buzzwords. No "unlock your potential" energy.
Length: 3 sentences max.
The main problem it solves: teams are paying for 40 tools but only
using 12 of them.
The first returns something generic enough to describe any app on earth. The second returns something you could ship.
That's not a Claude capability difference. It's an instructions difference.
The mental shift that makes everything click: Claude is a brilliant contractor who just started today. They're genuinely skilled. But they have zero context about your project, your standards, your audience, or what "good" means to you. Whatever you don't tell them, they'll invent.
Vague brief → generic work. Specific brief → the thing you actually wanted.