Second Brain, AI, and Obsidian: What I Explained to an Intern
Reflections on a discussion about Obsidian, second brains, and AI — and the confusion between useful tools, wow effects, and real effectiveness.
A conversation with an intern
While talking with an intern (Rodrigues), we stumbled onto a topic that has become very common in the AI ecosystem: the famous “second brain.”
The idea is appealing:
- centralize all your knowledge;
- connect information together;
- use tools as augmented memory bases;
- and, of course, connect everything to artificial intelligence.
In this context, tools like Obsidian often come up in discussions.
And that’s where the conversation took an interesting turn.
Obsidian is not magically useful for AI
I explained something quite simple to him:
Obsidian has, in itself, no direct interest for an AI.
An LLM does not “see” a knowledge graph.
It only sees text.
What matters to it is not the visualization of links, but:
- the structure of documents;
- the quality of content;
- the consistency of information;
- and the way it is retrieved (RAG, embeddings, etc.).
The graph, however elegant, is essentially a human tool.
What the tool really brings
On the other hand, Obsidian has two major benefits… but for humans.
1. Understand what you know (and what you don’t)
Link visualization allows you to:
- see connections between ideas;
- identify well-documented areas;
- spot isolated elements;
- detect “gaps” in knowledge.
And these isolated elements are important.
Because in an AI system, what is not connected often has little chance of being properly exploited.
2. Build a living knowledge base
Obsidian also helps produce a markdown note base:
- interconnected;
- readable;
- maintainable;
- scalable.
But this logic can exist without Obsidian.
A simple well-structured file system, with clear conventions, can produce exactly the same result on the AI side.
The most common misunderstanding
What I tried to make him understand is a fairly common misunderstanding:
“If it is well structured for humans, then it is automatically exploitable by an AI.”
In reality, that is not true.
An AI does not exploit a visualization.
It exploits a textual structure and search mechanisms.
And above all:
poorly structured knowledge remains poorly structured, even if it is beautifully represented in a graph.
The real issue is not the tool
As the discussion went on, we always reached the same conclusion in this type of project:
the problem is not Obsidian, nor the LLM, nor the architecture.
The problem is the actual structuring of knowledge.
Who writes what?
How is it validated?
How is it kept up to date?
What do we do with isolated information?
That is where AI projects succeed… or fail.
A broader drift: the wow effect
This discussion also reminded me of something broader.
In the current AI ecosystem, a lot of content is built around a specific goal:
- impress;
- oversimplify;
- show a spectacular demo;
- generate buzz.
This is not necessarily ill-intentioned.
But it creates a drift:
we often confuse what is impressive with what is effective.
A visually impressive graph is not necessarily a good knowledge base.
An AI agent that “speaks well” is not necessarily useful in production.
A series to come
This reflection around Obsidian and the second brain is really just an entry point.
It opens onto a much broader problem:
- information overload in AI;
- the difficulty of sorting technical content;
- the role of influencers and “wow” content;
- and the gap between demonstration and operational reality.
In upcoming articles, I will come back to this tension between:
what attracts attention and what actually works in AI projects.
Conclusion
Obsidian is not an AI tool.
It is a tool for human understanding.
And that is probably where the real issue lies.
In a world saturated with content, demos, and technical promises, the challenge is no longer just to build intelligent systems.
It is to know how to distinguish what is truly useful… from what is simply impressive.