The reflection blurs

I have spent hundreds of hours in conversation with AI over the past year — discussing the architecture of my research project, the philosophy of what I am building, my plans to move to another country. And recently I caught myself remembering almost nothing concrete from those conversations. Aletheia, my main research, is in a phase of deep engineering work. I am preparing to relocate. And in parallel I am looking for a second project that could financially support the work on Aletheia. I discuss all of these topics with AI often and at length. For conversations of this kind, it is the only interlocutor available to me.

This sheer volume of communication led me to an uncomfortable conclusion. Yes, I can state the general concept of the project. I can describe the architectural details and a few philosophical constants. I can even draw up a list of what needs to be done to relocate. But none of it adds up to a whole — not for me, not for the person I am talking to, not even for the AI itself. It has context limits, and in an abundance of information it begins to get confused. Just as I do. The skill of critical thinking atrophies when you delegate it to a machine. I feel this, and I resist it on instinct — I do not trust my conclusions to a machine, I want to think for myself. But the result of that resistance is that I have no external structure for my thoughts at all, and the context gets lost a different way.

I went looking for a solution on the market. In April 2026 Andrej Karpathy published the idea of an “LLM Wiki” — a concept where the AI reads your sources and compiles a knowledge base for you. The idea went viral, and implementations appeared. But they all embody a principle I find philosophically unacceptable: the AI does the work of thinking instead of you. These tools do not cure the problem — they make it worse, accelerating the atrophy of the very skill they touch. The metric becomes the volume of accumulated content. Not its quality. Not whether you ever return to a conclusion. Just quantity.

That is how the idea for a project — working name Kasane — was born. Its principle is the opposite of the mainstream: the AI proposes, the human decides. Every conclusion reached in dialogue with the AI is something you can accept or throw away. Every place in your document you choose yourself — where a new thought goes, what gets left out. The AI helps you structure. It does not decide for you what matters. The result is a document — but not a static archive. A living structure that changes as your understanding changes. The project develops, you move country, the world shifts, and the document evolves under the pressure of your own thinking. The history of changes is kept. You can watch how an idea unfolded.

Notion, Obsidian, Apple Notes — they all ask you to actively keep notes. Some people do this naturally. I do not. Every attempt to force myself to write in those tools failed — not from laziness, but from a mismatch with my real workflow. I think in conversation, not in a monologue in front of a blank page. And I suspect this is common. We know how to think through dialogue. We are bad at keeping notes with discipline. These are different skills. To this is added the specificity of Aletheia: I discuss it a great deal, and often, because there is no one else to discuss it with. It is already a long argument with myself, and Kasane settles it more naturally than any attempt to force myself into Notion or Obsidian.

To close, I want to say that this is the first entry in Aesleif’s notes, and I hope not the last — because Aesleif and Aletheia are a reflection of me, and a reflection blurs. As do I.