Keep AI context in Markdown, not in the chat

  • ai
  • tools
  • second-brain

I moved from OpenAI to Claude. Last week I started moving some of my setup from Claude to Grok. Each time, I had to bring my context with me, and the first time I tried, a lot of it did not arrive.

The imports were not empty. They were missing depth. Explanations I had given, preferences I had already set, and the depth behind each note were gone. I was largely starting from scratch, because so much of my context lived inside a conversation rather than in memory or a file.

That is a bad place to keep anything you might need again.

A chat is useful while you are in it. It is also tied to the product that hosts it. When you leave, you depend on that product’s export, and on how well the next product can read it. I do not want my working context to depend on that going well.

I also think AI is moving up the stack. Fintech is built on top of core banking infrastructure, and I expect AI to grow some kind of orchestration layer that sits above the individual models and tools. When that happens, I want my context to be ready to move with me rather than locked inside one product.

So I write the things I want to keep in plain Markdown files: my background, my decisions and how I like to work. The files are the source of truth, and the chat reads them. If I change tools again, I am not starting from a partial import and an empty prompt.

The idea isn’t new for me. During COVID, my brother and I got into Roam Research, Obsidian, bidirectional linking and the idea of a second brain. He kept his going. I didn’t follow through as well. This is my chance to fix that.

I am not saying this makes every conversation portable. A thread has a texture that a note does not. What I am protecting is the material I would otherwise have to retype: who I am, what I am working on, the calls I have already made, and the skills I am building into my AI agents.

If it still matters after the conversation ends, it belongs in a file, not in the thread.