Today, while cleaning out files on my computer and looking at everything I had accumulated over the last three years, I made a major decision: I deleted all the input material in my "second brain" archive. Everything that was not my own original output, plus anything generated purely by AI, is gone.
By coincidence, I happened to read an essay by Joan Westenberg today called I DELETED MY SECOND BRAIN. What he described felt almost perfectly aligned with what I felt the moment I hit delete:
Erased in seconds. What followed: Relief.
Why Delete It All?
Ever since I first discovered Obsidian three years ago, I have been endlessly tinkering with personal knowledge systems. At first I organized everything by hand. Later I started using AI to help summarize and structure things. Over time, I accumulated a massive amount of material.
And then I slowly realized something obvious: once the pile gets big enough, you never actually look at most of it again. Maybe 90 percent of what I saved was read once, at the moment I clipped it, and never opened afterward.
What we once believed was "knowledge management" had quietly turned into a mental burden. As Westenberg put it:
But over time, my second brain became a mausoleum. A dusty collection of old selves, old interests, old compulsions... Instead of accelerating my thinking, it began to replace it. Instead of aiding memory, it froze my curiosity into static categories.
Because in the AI era, I no longer think early-stage collection and categorization matter as much as they used to. Retrieval is now unbelievably powerful. AI can search the internet, and AI agents like Cursor or Claude can search through your local folders just as easily. The organizational labor we used to do in advance can increasingly be delegated.
That hit me especially hard a few days ago while gathering material for a podcast episode about pop music. You can tell an agent what you need, and it can go search official sites, dig through reviews from respected critics, even write a scraper to scan Bilibili or Xiaohongshu for public reactions, then bundle it all together for you. It can even pre-organize the material so you do not have to read everything in full. You just inspect the result and decide what to do next.
That gave me a new way to think about the whole archive: this is a burn-after-reading era.
Carry the Key, Not the Warehouse
Maybe we do not need to save everything anymore. Once you have understood something, you can let it go. What you really need is the core concept, the name of the thing, the intellectual handle. When I study behavioral economics, for example, I do not need to preserve every article I read. I just need to know what "loss aversion" is, what "sunk cost" means, and where those ideas connect to other things.
What matters more is the act of connection.
Say I read an article about how sleep affects memory and comprehension. It may contain a few terms, a few experiments, a few findings, but the essence of it could probably be reduced to fewer than fifty words. If you can connect those fifty words properly, you already understand the piece. And when you need it later, you can pull up the original article through the saved link, pair it with your short note, and let AI help generate a full reading report from there.
You do not need to organize everything into perfection ahead of time. You just need the fifty-word key. When the moment comes, you open the door and retrieve what matters.
Otherwise we fall into a trap of pretending to think:
The more my system grew, the more I deferred the work of thought to some future self who would sort, tag, distill, and extract the gold. That self never arrived.
From Librarian to Search Engine
In the past, the human mind had to behave more like an encyclopedia. You needed to retain a great deal of concrete knowledge because search was weak and reorganization was expensive. But now, if you know which ideas should be placed in relation to each other, AI can do the assembly almost instantly.
That means our cognitive model has to change. We need to move from being a "librarian" of stored knowledge to becoming something more like a "search engine."
You should function more like a sitemap: a system of indexes and retrieval paths, connected to many sources of information, always ready for a fresh search. Let AI handle everything AI is good at. Keep only the parts it cannot do for yourself.
That is the direction I am moving in now. When I find a good article, I no longer use the Obsidian web clipper to save huge chunks of it. I do not even need folders anymore. One file is enough: write down the article title, paste the link, summarize it in one sentence, add a few keywords, and stop there.
To me, that is a more sensible way to use AI. AI excels at system-building. It is great at expanding from one to one hundred. But it is much worse at zero to one, because its thinking is inherently rigid. What matters more now is connection, cross-domain synthesis, and interdisciplinary movement. Only human beings can make those strange and meaningful links.
And once you give AI that original cross-disciplinary direction, it can follow your lead and produce an enormous amount of output. Human beings only need to do the opening move and the closing move: begin the thought, then edit the result so it becomes more coherent, more alive, and more human.
Becoming the Boss of Knowledge
If I had to reduce today’s realization to one sentence, it would be this: we have entered a burn-after-reading age. People no longer need to preserve every input. What matters now is output and thought.
We used to work for knowledge. We had to memorize and store its contents in order to use it. But now that AI exists, our role has changed. We are no longer the laborers. We are the managers. AI has become the employee that works on knowledge for us.
All we need is the moment when an unexpected thought appears: "Could these two things belong together?" Then we find them, throw them to AI, and ask: "I think there’s a connection here. What do you see?" If the result is good, we keep it. If it is not, we keep thinking. That feels like a healthier, more efficient, less wasteful way to produce work.
Pure knowledge intake might once have soothed our anxiety. But now, apart from making it easier to avoid the harder work of output, it does almost nothing.
Just like the decision Westenberg reaches at the end of his essay:
I don’t want to manage knowledge. I want to live it.
In this new era, let AI do what AI should do, and let our minds do what only our minds can do.