Am I getting as much out of AI as I could?
Most of this course has been about where AI goes wrong. This lesson is about the opposite problem.
In September 2026, Ethan Mollick, a professor at Wharton who writes a lot about AI at work, described a big gap:
"The capability overhang, the gap between what these models can do and what almost anyone is doing with them, is an opportunity."
Most people use a very capable tool for a small set of everyday jobs: rewording emails, summarising, quick questions.
In the same piece, Mollick's examples include turning a 1977 text adventure game into a playable 3D game, and rebuilding a famous writer's library in 3D from photos and video.
You don't need to do anything like that. The point is that the ceiling is far higher than most people's daily use.
Anthropic looked at how Claude is used. People who had been using it for six months or more had a 10% higher success rate in their conversations, even after accounting for what tasks they did and where they lived.
They also took on more complex work, and worked with it in more back-and-forth, rather than just handing it orders.
People get better at it with practice. It's a skill.
Mollick lists four things people bring that AI doesn't:
Everything earlier in this course relies on the first three. This lesson is about the fourth.
Models change quickly. Something it was bad at a year ago may be easy now. Something you gave up on after one bad try may just have needed a better brief (see "AI can't read your mind").
The person who tried AI once in 2024 and decided it was useless is working from an old picture.
Be ambitious about what you try, and careful about what you trust.
Once a week, give it a task you'd normally assume is too hard. Brief it properly.
Ask it what it could do with what you have. "Here's my spreadsheet. What are five useful things you could do with it?"
Keep a short list of things it failed at, and try them again when a new model comes out.