As someone who has been building a second brain on Obsidian since 2020, the hype on AI’s role on this topic needs a reality check.
The current pitch is that AI can now do the tending for you. It captures, tags, links, summarises and resurfaces. Point it at your vault and it will tell you what you think. I use some of this myself, but most of the pitch misreads what a second brain is for.
Yama’s chariot
The Katha Upanishad, from the same texts that gave me neti neti, has a better model than any productivity thread. Yama, teaching the boy Nachiketa, describes a person as a chariot. The Self is the rider and the body is the chariot. The intellect is the charioteer, the mind is the reins, and the senses are the horses, running over the roads of everything they perceive. When the charioteer lacks understanding and the reins hang loose, the horses run wild. When the charioteer holds the reins firm, the rider reaches the end of the road (Katha Upanishad 1.3.3-9).
Map a second brain onto that. Every feed, highlight and saved article opens another road for the horses. AI makes the horses faster: more captured, more connected, more resurfaced, at a pace no person can match. Nothing in that stack strengthens the charioteer or tightens the reins. Add horsepower to a chariot whose driver is asleep and you get dragged further, faster, in directions you never chose.
The moment after a signal lands
The moment that matters is small. A signal lands: a remark in a meeting, a line in a book, a number on a dashboard. Before you have a thought about it, something in you has already reacted. Soon, Brass, Heinze and Haynes (Nature Neuroscience, 2008) found brain activity that predicted a simple left-or-right choice several seconds before people reported deciding. My read of that work is that agency lies less in when an impulse starts and more in whether you notice it and can override it. That noticing is the charioteer’s job, and no model has access to it.
So the work after a signal is internal. What changed in how I see this? Why did it land on me this way? What did I feel in the first moment, before I had a story for it? If you skip those questions, the note you file only records where the horses went. AI will link it neatly to forty other notes, and you will drive no better than before, because the mind returns to its natural state unless something inside it has changed.
Compounding needs processed deposits
I have argued that notes compound like interest. They do when the deposit is something you have processed. A highlight you never rewrote in your own words is borrowed money. Compounding on it gives you a bigger pile of other people’s conclusions, sorted by an algorithm that can’t tell which of them changed you.
Slow gardening
The gardening I would defend is deliberately slow:
- Write each note in your own words, even when an AI summary would be faster. Writing is how you name what the signal did to you.
- Sleep on anything charged before you file it. A first reaction is a lossy compression of an incomplete view.
- Hold a weekly review where you reread old notes and ask whether you still believe them.
- Let AI hold the tongs: formatting, retrieval, finding the note you forgot you wrote. Keep the judgment of what mattered for yourself.
Writing this way carves channels in advance. When the next surge of reaction arrives, it flows through a shape you built on purpose and stops ricocheting off edges you never examined. In practice, that is what holding the reins means.
AI gives a second brain more horses than anyone has ever had. Whether they take you anywhere depends on the hours you spend holding the reins.