Journal › Where to start
What should I use AI for first?
The short answer
Don't start with a task. Start with your own thinking: use the AI to challenge what you assume it can do, and what you assume you're capable of, before you point it at anything. The usual advice is to pick a small, low-risk job off your to-do list — and that job quietly becomes your ceiling, because everything you build afterwards is shaped by what you already believed was possible.
Everyone tells you to start small. Nobody tells you what starting small costs you.
What I actually did
I didn't start with a job. I started with the thinking. I've got big ideas and big ambition, so the first thing I wanted to understand was how this thing reasons — and what I could do to make it reason better. Where it's strong. Where it's confidently wrong. Then I turned it round and had it argue with me about what I actually wanted from it, and about the limits I'd assumed without noticing I'd assumed them. That conversation moved my own line for what was possible, and everything I've built since sits on the far side of it. If I'd opened with "summarise my emails", I'd have ended up with a very good email summariser and stopped there.
What you can check
- Ask it to argue against something you've already decided. If it just agrees with you, you've built an assistant, not a thinking partner — and an assistant will never widen what you think is possible.
- Write down how it reasons, not just what it produces. I keep one file per thinking method, each with where the idea actually came from and the strongest published argument against it. It's the difference between using a tool and knowing when to distrust it.
- Test the ceiling deliberately: hand it something you assume is too big. In one evening mine read roughly 23,000 lines of original source material and built a working method for my workshop out of it — then a rival AI, brought in blind, found eleven faults in that work before I relied on any of it.