One month with my second brain — what I got wrong and how I fixed it

I built my second brain in a weekend and immediately started using it wrong.

Not wrong in a technical sense. The setup worked. Claude was ingesting sources, building wiki pages, connecting everything. The system was running.

I just had no idea what was going on inside it.

The mistake I was making

I was treating my second brain like a magic oracle. Dump content in, Claude organizes it, knowledge appears. I'd drop a raw source into the folder, tell Claude to ingest it, and trust that whatever came out the other side was correct and useful and in the right place.

I wasn't checking. I wasn't deciding. I was hoping.

The problem with that approach is obvious in retrospect: Claude is only as good as you make it. If you don't understand the system you're building, you can't tell when it's not working. And if you don't know where things are supposed to live, the connections Claude makes might be technically correct and still completely useless to you.

I was outsourcing the thinking. And the thinking is the whole point.

The feeling that something was off

I'd open the vault and feel vaguely lost. Not lost in a "this is broken" way — more like opening a room that someone else organized. Everything was there. I just didn't know where anything was.

I didn't understand how my files were being linked together. I didn't know why certain pages existed in certain folders. I'd open the graph view expecting it to click into place — expecting to look at it and just get it — and it never quite did. It sat there, technically accurate, meaning nothing to me.

The thing is, none of that was Claude's fault. A vault only feels like yours once you've made the decisions — what goes where, how things connect, what each folder is actually for.

I hadn't made any of them. I was just trusting the AI to figure it out.

What we actually did about it

One session, I stopped adding content and started asking a different question: what is actually in here, and does it make sense?

Me and Claude did a proper analysis. We went through every folder and asked: does this name make sense? Do I know immediately what goes here? Is this connected to the right things? When I navigate from one page, can I get to what I actually need next?

The answers were uncomfortable.

Entities/ was supposed to hold people and companies. But "entities" is such a vague word that nothing had an obvious home. Every time I opened it I'd have to think — is this an entity? What even is an entity? I'd avoid the decision entirely and leave the file somewhere else.

Synthesis/ had the same problem. I didn't know what it was for, so I never consciously used it.

My projects and my clients were in the same folder. My own product sitting right next to a client's name, like they're the same kind of thing. They're not. One is mine. The other is income. That confusion was subtle but it affected how I felt every time I opened the folder.

And the links — I had no strategy for them. Claude was connecting things that were technically related, but I hadn't decided what should be connected or why. So the graph view was noise.

What I changed

First, I renamed everything to words I actually use and understand.

Entities/ became People/ and Companies/. Now it's obvious. Matilde goes in People. My clients go in Companies. Zero thinking required.

Synthesis/ became Insights/. These are files Claude built by thinking across my raw sources — deal structures, market signals, strategy overviews. They're conclusions. Insights is what they are.

I separated my clients from my own projects. 04 Projects/ is mine only — Loomi, MindBuddy, Innerloop, the personal website. 08 Clients/ is client work. Opening Projects now feels like opening something that belongs to me.

Then we designed the linking strategy together — five rules I actually understand and agreed to. Not Claude's rules. Mine. Things like: every career concept links to my page, everything that supports the visa links to the visa strategy, concepts link to the projects they most directly inform. Now when I open a page, the related links actually make sense to me.

And I started getting involved in the decisions. When a new source gets ingested, I review what was built. When a new folder is created, I confirm the name works for me. When Claude proposes how something should be connected, I check whether that connection is useful.

What I actually learned

The version I had before — I'd built it but I didn't own it. There's a difference. Claude had organized it and I trusted that was enough. It wasn't.

The version I have now, I actually understand. I know what's in it. I know how it's connected. When I open the graph view, it makes sense to me.

That's the version that actually helps.

Mariana