#12 How to start
People keep asking me how to start. Not how to use Claude. More and more people know that. But how to build something like Pippi. A system that actually knows how they think, that remembers between sessions, that pushes back instead of just agreeing.
Pippi took weeks and it started from the inside, from my own patterns and friction points. I did not start with a framework. The framework became visible through what I built around my own thinking. In my previous post I shared how Pippi became Alva when Marit went through my methodology and built her own.
You can describe your job title, your preferences, maybe even your communication style. But you can also go deeper and describe how your mind actually works. The gap between how you describe yourself and how you actually think is where most AI setups fall short today. The system does what you ask. But it does not know how you think. And that changes everything about what it gives you back.
I designed Mind Print to help close that gap.
It is a set of questions and they are maybe not the questions you expect. Not what do you do or what are your strengths. More like, when do you do your best thinking and what does that look like from the outside? What does AI do that consistently misses the point? What kind of pushback makes you sharper and what kind just wastes your time?
Some of them will feel obvious until you sit with them. One of them asks what you are good at that you barely notice yourself doing. Not because it is hard, but because the answer is something so automatic you never thought of it as a skill. That is exactly the kind of thing AI needs to know about you and that you would never think to write down.
When you finish, Mind Print gives you two things. A short reflection on your cognitive patterns, specific enough that you will probably feel a small shock of recognition. And a starter CLAUDE.md. A document written for AI about how you think, how you work, what to never do.
It is a first draft. It will be wrong in places. That is fine, because the errors are information.
When AI does something that feels off, most people correct the output and move on. But if you have a CLAUDE.md you can ask a different question. What is missing from the instructions that led to this? Usually the answer is something true about how you work that you never thought to write down because it felt too obvious.
The gap between what you meant and what AI did is a direct line to your own blind spots. That is not a debugging process. That is a thinking practice.
When AI gets it right, you notice why and reinforce it. When it gets it wrong, you trace the failure back to an assumption you did not know you were making. Over time the document stops being a configuration file and starts being something closer to a map of how your mind works. And the map keeps getting more accurate because the territory keeps revealing itself.
And then something happens that I did not expect when I started. You change too. When an AI that knows your cognitive patterns pushes back on a decision, it forces you to rethink assumptions you did not know you were making. When you correct it, you are not just improving the system. You are teaching yourself something at the same time. The more you pay attention, the harder it becomes to give your thinking away without noticing.
I wrote a guide for what to do after Mind Print. If you have never used Claude Code, it starts from zero. Setup, first conversation, how to give your AI an identity and memory and skills. If you already have something built, it starts from the cognitive layer most setups are missing. Both tracks end in the same place, with testing whether your AI actually knows you and what to do when it does not.
Start with Mind Print. See what it surfaces.
Then follow the guide and start paying attention to the gaps.
That is where the work is. Not the setup. Not the commands. The practice of noticing how you think and using it to make better decisions.