Build a Second Brain You Actually Own

By Dan Sanchez · Episode published

AI models keep leapfrogging each other. Build a portable knowledge base with local-first notes and shared agent access so your context stays yours when the winning model changes.

Get the next practical AI marketing episode wherever you listen.

Apple PodcastsSpotifyYouTube

The AI model you use today may not be the model you use next year. ChatGPT, Claude, and Gemini are all improving quickly, and the lead keeps changing. That makes one thing more valuable than a killer prompt: context.

If your best ideas, decisions, transcripts, and working notes are trapped inside one AI app, switching models means starting over. I want a setup where the model can change while my information stays with me.

Put your notes before the app

The foundation I am exploring is Obsidian, a local-first notes app that stores information as ordinary Markdown files. It feels like a note-taking application, but underneath it is a folder on your computer. Notes are plain text. Images and other files live alongside them. You can sync that folder across devices, or keep it local.

That detail changes the ownership model. If I stop liking Obsidian, I can move to another interface without having to export my life from a proprietary database. The notes are still files that another tool can read. The app is an interface over the knowledge, rather than the place where the knowledge is held hostage.

This is the idea behind “notes before apps.” It is useful even without AI because a folder of linked Markdown files is easier to search, move, back up, and reuse than scattered documents in several services. It becomes much more powerful when an agent can navigate the folder for you.

Give several agents the same context

ChatGPT Codex and Claude Code are built to work through repositories. They expect information to be spread across folders and files, so they are good at inspecting a body of material, finding the relevant pieces, and taking an action. They do not have to be used only for building websites.

Connect one of these agents to an Obsidian vault and you can ask it to find the notes related to an idea, organize a crowded folder, create a new note, or turn a collection of transcripts into an outline. If the next model is better at writing, use that model. If another is better at a different task, give it access to the same files. Your context survives the switch.

The important work still belongs to you. You need to decide what gets saved, approve major reorganizations, and check what an agent changes. The point is to make your knowledge portable and available, not to give an unreviewed process permission to rearrange your thinking.

Use deep research with a defined source boundary

The same principle applies to research. ChatGPT’s Deep Research is much more useful when you tell it where to look. You can require specific websites, emphasize trusted sources, or connect resources such as a Google Drive. That lets you ask a research agent to work from your own material instead of blending it with whatever happens to rank on the open web.

I tested this by restricting research to my own sites and asking for an outline of a book about what it means to be an AI-driven marketer. It took time to read the material, then returned chapters, recurring ideas, supporting evidence, and links back to the source. That is a very different result from asking for a generic book about AI marketing.

For marketers, this creates a practical workflow: collect customer language, past content, research, frameworks, and decisions in a place you control. Then use whichever model is best today to retrieve and reshape that context. The model may change. Your accumulated thinking does not have to.

The future-proof move is simple: own the files, keep the structure portable, and let the agents compete on top of your knowledge base. That gives you flexibility without making every new model release a reason to rebuild your brain.