Is the Claude and Obsidian Second Brain Trend Missing the Point?

Lately I keep seeing posts about using Claude and Obsidian to build a second brain. Some teach you to feed everything to Claude and have it generate notes and links between them, producing a beautiful graph. One set of documents might get 500 nodes and 124,750 edges from Claude (that’s a complete graph; to see how that number is calculated, you’re welcome to join the AIC-504 Graph-Based Learning class at CMKL).

I’ve used Obsidian since around 2021. Before that I followed Tiago Forte and his Building a Second Brain and PARA since 2020, and before that I used plain OneNote. I’d like to invite you to think for a moment about whether what’s trending may be missing the point of having a second brain.

First, we need to understand that the simple principle of a second brain is to let us offload information or memories that don’t need to stay in our heads, so the brain has room for synthesis and creative work. What these second brains should do is help us think, not think for us.

From five years of experience, I’ve found that the most valuable moment in Obsidian is not when I’m writing a note and saving it. It’s when I open a note and type a wikilink ([[ ]]) while thinking of something, then glance at the graph view beside it and it shows me old notes I wrote that I’d sometimes forgotten. That lets me begin to see clusters of thoughts slowly forming (for example, I was writing about making plant accessories to sell, and it linked to regenerative farming).

This kind of moment is what Tiago calls serendipity: two ideas that seem unrelated meeting in our heads and producing a new insight (another person who talks about this often is David Perell). But this kind of moment can happen only when we write the notes ourselves, because writing forces us to process the information, organize our thoughts, cut what’s unimportant, keep what’s important, and decide what to link it to.

Sometimes I create empty wikilinks because I think I’ll come back and think more about the concept later. It makes me use a bit of my own cognitive work, to recognize that concepts like these are related.

The problem with the current trend is that having an LLM generate both the notes and the links removes this step entirely. What’s left in Obsidian is a beautiful artifact, a graph view full of lines, but no cognitive residue in our heads at all.

It’s like a second brain that isn’t really connected to our first brain.

This doesn’t mean LLMs have no place in this workflow. I use Claude to help with many things myself, including keeping various contexts in Claude through these very notes. But I’d like everyone to try exploring the concept of building a real second brain. Then we get to see AI augmenting the thinking process, instead of thinking 100% for us.

I’ll make a tutorial later on thinking with graphs, both for research and for note-taking.

One last thing. Many people optimize the graph view by trying to make every node connect with many others. Some use an LLM to fill in links, believing that the more connected it is, the better the second brain.

In graph theory there’s a test called the 1-Weisfeiler-Lehman test, used to tell whether two graphs have different structures. It leads to an interesting insight: a fully-connected graph actually gives less information than a sparse graph with clear structure.

In plain language: a graph where every node connects to every other node, and a graph with no links at all, give the same amount of structural information.

This has direct implications for how we design our knowledge graphs, but I’ll save it for the next post. It would get long, haha.