I have been journaling for six years. Not because I had a system. Not because I read a productivity book and decided to optimise myself. I started because I needed somewhere to put things — thoughts that were too unfinished for a conversation, observations I didn’t want to lose, the occasional reckoning with how I was actually doing versus how I said I was doing.
I didn’t call it a second brain. At first it was a GTD thing implemented on Evernote. After that I moved it to UpNote.
Recently I came across an idea by AI researcher Andrej Karpathy – about feeding your personal files into an LLM and letting it build a structured knowledge base from them. A persistent, synthesised layer that sits between your raw notes and your questions. Not just search. Actual synthesis. The LLM reads everything, builds connections, flags contradictions, and maintains a wiki that compounds over time as you add to it.
I tried it with my own journals. What came back was genuinely strange to read. Six years of thinking, compressed and cross-referenced. Patterns I hadn’t noticed. Things I had apparently decided, then un-decided, then decided again. Evidence of change I hadn’t consciously registered. It was like reading a biography of yourself written by someone who had no reason to be kind about it.
But here is the thing that stayed with me afterwards.
This only works if you have a corpus. You cannot build a second brain from nothing. The LLM is extraordinarily good at synthesis — but synthesis requires material. Feed it nothing and it will return nothing, dressed up beautifully in markdown. Garbage-In-Garbage-Out.
And most people have been thinking their whole lives and have almost nothing to show for it. Not because they are incurious or unreflective, but because the thinking evaporated. The commute thought. The 2am realisation. The meeting observation that was sharp for thirty seconds and then gone. We generate enormous amounts of insight and retain almost none of it in any retrievable form.
The AI moment changes the calculus on this in a way that I don’t think has fully landed yet.
Previously, capturing notes was only as useful as your ability to manually retrieve and connect them. Which is hard. So most people stopped, or never started. The effort of maintaining a system exceeded the visible return. Tiago Forte has spent years arguing against this — that the platform doesn’t matter, that the practice matters, that any consistent capture habit beats the perfect system you never implement. He is right. But even he was asking people to do the retrieval work themselves.
Now you don’t have to. The LLM does the retrieval, the synthesis, the connection-making. Your job is just to have written something down.
So the argument I want to make is this: start now, and go back.
Start now — not with a system, not with tags and templates and a Notion database that takes three weeks to design. Just write. Capture the thought. Put a date on it. Keep it somewhere that exports to plain text. That is enough. The structure can come later; the LLM can impose a surprising amount of it retroactively. What it cannot do is recover the thought you didn’t write down.
And go back — because you probably have more than you think. Old emails. Work documents. A blog nobody reads. Half-finished notes from a course you took in 2019. Voice memos you forgot to delete. None of it is perfectly formatted. None of it needs to be. Gather it, normalise it to the extent you can, and it becomes a corpus. A corpus becomes something an LLM can work with. Something an LLM can work with becomes, finally, the second brain that the productivity books always promised but couldn’t quite deliver on their own.
The real insight from Karpathy’s idea is not the technology. It is that the value of your thinking is no longer limited by your ability to organise it yourself.
It was always worth writing things down. It is just that now, there is a much better reason to have done it.
Leave a Reply