The second-brain trap

I have a folder on my laptop called graveyard. It is where good ideas go to be politely forgotten. It is, without question, the most expensive folder I own.

I started it five years ago, when I read my first book about a "second brain." The book was good. The pitch was irresistible: capture everything, link it together, let your past self do your present self's thinking, retire to a cabin somewhere and live off the compounding interest of your notes. I bought the book. I bought the app. I bought, eventually, three different apps, because I kept noticing that I had outgrown each one in a way that was probably the app's fault. I imported my old Evernote. I imported my Things tasks. I imported my browser bookmarks. I had, by the end of one frantic weekend, eleven thousand notes.

I have, in the half-decade since, opened fewer than two hundred of them.

This is the second-brain trap, and it is not a personal failing. It is the predictable outcome of a system that was sold as the solution to a thinking problem when in fact it is, structurally, a hoarding system disguised as a thinking system.[1][1]The same way a chest freezer in the garage is a hoarding system disguised as a meal-prep system. The freezer is not the problem. The freezer is, in fact, working perfectly. The problem is the relationship between the freezer and dinner.[1]The same way a chest freezer in the garage is a hoarding system disguised as a meal-prep system. The freezer is not the problem. The freezer is, in fact, working perfectly. The problem is the relationship between the freezer and dinner. This essay is about how to tell the difference, and how to build the kind that actually pays you back.

FIG · B.07SECOND-BRAIN PIPELINE

The receipts

I want to be clear: I am not anti-notes. I take a lot of notes. I take notes during meetings, after meetings, before meetings, in the shower, in the woods, at 2 a.m. when I should be sleeping. Notes are how I think. The case I am making is much narrower:

The case is that the system most second-brain advocates sell — capture everything, tag everything, link everything, trust the future-you to handle retrieval — is structurally broken in a way that nobody who is also selling you a course on it is incentivized to admit. The system fails for a reason. The reason is that it confuses input with output, and the entire economy of attention runs on output.

Here is the receipt that finally turned me. Earlier this year I opened my notes vault for the first time in three months. I searched for a topic I had researched, deeply, in 2024 — about half a megabyte of careful, linked, tagged notes. I needed a particular fact from that research for an email I was writing.

I could not find the fact. I scrolled. I clicked through eight Markdown files. I tried four different searches. I gave up after eleven minutes and just used Google.

The search took six seconds.

I had spent — conservatively — sixty hours researching that topic, and the aggregate retrieval value of the resulting notes, on the only occasion I needed them, was negative one minute. They cost me eleven minutes of futile searching that Google would have answered in six seconds.

If you have a second brain, run this experiment yourself before you finish this essay. Pick a topic from a year ago. Try to retrieve a specific fact. Time it. Then time the same retrieval from Google. Be honest about the result. The honest result is the entire essay; the rest of this is just me explaining why.

A note-taking system you can't beat Google with is not a thinking system. It is a feelings-of-thinking system, which is more expensive and worse.

What's actually broken

Three things, in increasing order of how unfashionable it is to say them.

The first is the capture obsession. The seductive promise of every second-brain system is that capture is cheap and retrieval is expensive, so we should optimize the cheap operation. The optimization works, in the trivial sense that you can now capture an article in two keystrokes. But "cheap to capture" turns out to be an asymptote — it bottoms out around 2018, after which every additional ergonomic improvement is just adding more inputs, not making them more useful. Meanwhile retrieval is not getting cheaper, because the cost of retrieval is dominated by how similar your stored notes are to each other, not how easy they were to capture. Eleven thousand undifferentiated bookmarks of "this looked interesting" are functionally a search problem that scales worse than the open web. You haven't built a second brain; you've built a worse Google with no PageRank.

The second is the linking fetish. The other half of the pitch is that you should link related notes, because the links will form an emergent graph that is your thinking. This sounds beautiful and is mostly false. The links you can manually maintain at human scale — say, a few thousand notes — are too sparse to be a real graph. The links you'd need to actually approximate a graph — say, every note linked to the eight most semantically similar notes — would require you to read every note every time you take a new one, which nobody does. So in practice you get a graph that looks like a constellation: a few rich clusters around topics you've worked hard on, vast empty space between them, and a long tail of notes with zero links that exist only because the system told you to capture them. The graph is not what's doing the work. The clusters are what's doing the work, and you didn't need a graph to make a cluster — you needed a folder.

The third is the disposability blindness. The deepest, weirdest failure mode of these systems is that they treat all notes as equally permanent. A fleeting thought you had at 11:14 a.m. on a Tuesday is filed in the same architecture as a paragraph you spent two hours crafting about the central mistake of your career. The system has no opinion about which is more important. The user is supposed to supply the opinion, via tags or stars or whatever. But the user is the same user whose attention is what we were trying to free up in the first place. We have not solved the curation problem; we have merely moved it from the moment of capture (where it was hard) to the moment of retrieval (where it is now also hard, but also late).

The three rules

After five years of building, abandoning, rebuilding, and finally mostly burning my own systems, I have arrived at three rules. They are not glamorous. They are also, as far as I can tell, the only rules that actually pay rent.

Rule one — the system has to cost you something at capture

The single biggest pathology of digital note-taking is that capture is too cheap. When capture costs a quarter-second, you capture twenty things a day; nineteen of them are noise. When capture costs ninety seconds — say, you have to write a sentence summarizing why you saved this thing — you capture three a day, and all three of them are signal. The summary is the filter. There is no other filter that works.

In practice this means: I do not save articles. I save one sentence about an article, in my own words, in a single text file. If the sentence is hard to write, the article wasn't worth saving. If the sentence is easy to write, I now have a note that retrieves on the idea, not the title.

This rule is the entire ballgame. The other two rules are about what to do once you've adopted it.

Rule two — the file is the artifact, not the database

The temptation in any second-brain system is to treat the system itself as the deliverable. You optimize the schema. You refactor the tags. You migrate from Notion to Obsidian to Logseq to Tana, each migration occupying a weekend and producing zero new ideas. This is not the use of the system; this is the worship of the system. The system is supposed to produce artifacts — essays, decisions, reports, memos, code, products. If at the end of a year your output is "a better-organized notes vault," you have spent a year building the freezer and forgetting to cook.

In practice this means: every quarter, I take the most useful thirty notes and turn them into something. A blog post. An internal memo. A decision document. A diagram. A talk. The notes are the raw material; the artifact is the output. The artifact is what gets retrieved later — by me, by other people, by search engines, by the hiring manager who Googles me in 2031. The notes themselves are, mostly, scaffolding to be discarded.

Rule three — the retrieval test, run quarterly, no exceptions

This is the rule that almost nobody implements, because it is the rule that exposes whether the rest of the system is working. Once a quarter, sit down and pretend you need to give a talk on a topic you haven't thought about in six months. Use only your notes. Time how long it takes to assemble the outline.

If the answer is "I couldn't, I had to use Google" — your system is broken. Not "needs tweaks." Broken. The capture isn't filtering. The structure isn't surfacing. You are, at best, paying the cost of a notes system to feel productive.

If the answer is "ten minutes and the outline is better than the one I'd have written from scratch" — congratulations. You are in the rare territory where the second brain is doing what it claimed it would. Most people never visit this territory. Most people have, like me, a graveyard.

What this looks like, in practice, for me, in May 2026

For what it's worth, here is what survived the burning. I keep one daily file (text). I keep one project file per active project (text). I keep a single ideas.md (text) for things I might want to write about later — and the rule is that each idea has to fit on one line. Ideas that don't fit on one line are ideas that aren't ready, and ideas that aren't ready don't make the file.

That's the whole thing. There are no tags. There is no graph. There are no apps with proprietary file formats. There is no daily review. There is, every quarter, a thirty-minute session where I read the previous quarter's daily files in chronological order and write a single page summarizing what I learned.

The page goes into a folder called quarterly. That folder is, in some real sense, my second brain. It is six pages long.

The graveyard, meanwhile, is still there. I haven't deleted it. I keep it around as a memorial — to the idea that capture is the same thing as thinking, to the eleven thousand notes I will never read, to the years I spent organizing the freezer instead of cooking the meal.

// DIAGNOSTICHONESTY CHECK

In the last 30 days, how many times did you successfully retrieve a useful insight from your notes?

12345

A coda about the people who sell these systems

Not all of them are charlatans. Some of them have built genuinely useful systems for their work — usually because their work happens to be making content about systems for organizing knowledge, which is a domain where the system happens to be the artifact, which is a domain that is not yours.

When you watch someone with thirty thousand interconnected notes give a talk about how their system saved their career, you are watching someone who has made the system the career. That's fine. It is not, however, advice. The right question to ask any guru of any productivity system is: what would your output look like with half the system? If the answer is "the same," you've found a real one. If the answer is a panicked recitation of why you can't do the work without the schema, you've found a hostage.

I have been the hostage. I am not the hostage anymore. The freezer is empty. I cook now.

It is, somehow, better.

▸ READ NEXT

curated by signal · not by algorithm

• build·2026-05-06·14 min

  Latency is a worldview

  Network engineering as a metaphor for how good thinkers structure their lives. Async, batched, queued, cached — and the one bad path that ruins all of them.

  read essay →
• teach·2026-05-04·4 min

  The system I actually use

  The full inventory: every tool, template, and ritual I run my work through. Not the version on the conference talk — the version that survives a Tuesday.

  read essay →
• mind·2026-05-07·16 min

  The cost of being available

  I once calculated, in a fit of late-night rigor, that I had been on call — in some informal sense — for roughly 4,200 of the previous 4,380 days. The number was wrong only in that it was too low.

  read essay →

STUDIO