# How to build a second brain that outlives the person who built it

> The CODE method in Tiago Forte's own words, then the assumption under all four steps: it is built for one reader. What to change when a team shares it.

Source: https://baalda.com/blog/how-to-build-a-second-brain
Site: Baalda, https://baalda.com
Published: 2026-10-13

**Short answer.** Build it in four steps: capture only what resonates, organise around your active projects, distill each note in layers, and express the result as something finished. Tiago Forte's CODE method is sound. It assumes one reader, your future self, so a team has to add a fifth constraint: every note legible to someone who was not there.

There is a four-step method for this and it works. It has also been restated so many times that the restatements have crowded out the one thing in it worth arguing with: every step is calibrated to a single reader, and that reader is you, later. If you are building a second brain so that you can find your own thinking again, follow the method as written. If you are building one so that a team can, there is a fifth requirement nobody puts in the list, and it decides whether the thing is still useful in two years.

## What is a second brain, actually

Tiago Forte, who popularised the term, defines it as "an external, centralized, digital repository for the things you learn and the resources from which they come". That is worth reading closely, because three of those words are doing the work. **External**, so it is not in your head. **Centralized**, so there is one of it. **Digital**, so it can be searched rather than remembered.

What the definition does not say is who it is for. Forte's own design question, from the piece where he introduced progressive summarization, is explicit about it: "How do I make what I'm consuming right now easily discoverable for my future self?" Your future self. One reader, with one person's taste and one person's memory of why a note exists.

That is not a criticism of the method. It is a description of its scope, and most people typing this query are inside that scope. It matters only at the point where a second person has to open the thing, which is the point Baalda, the app this site is for, was built for: one set of plain markdown files on your own disk that several people edit at once, with access set per folder and per file. Keep that aside until the end of this piece, because the method comes first and no tool substitutes for it.

## The four steps, and what each one is actually asking of you

The method is CODE: Capture, Organize, Distill, Express. Forte describes it as "a proven process for consistently turning the information you consume into creative output and concrete results." Here is what each step asks, in his terms rather than the usual paraphrase.

**Capture.** "Capture only the most important information", and the test for important is to "save anything that 'resonates' with you on an intuitive level". Not a rule, a feeling. This is deliberate: he argues elsewhere that an analytical checklist is too slow to survive contact with a tired person, so the criterion is intuition.

**Organize.** "The best way to organize your notes is to focus on your active projects", under the principle "Organize for actionability". In practice that means PARA, which Forte calls "a simple, comprehensive, yet extremely flexible system for organizing any type of digital information across any platform". Four folders, and his definitions of them are tighter than most summaries admit:

| Folder | Forte's definition |
|---|---|
| Projects | "short-term efforts (in your work or personal life) that you take on with a certain goal in mind" |
| Areas | "important parts of your work and life that require ongoing attention" |
| Resources | "a range of topics you're interested in and learning about" |
| Archives | "anything from the previous three categories that is no longer active, but you might want to save for future reference" |

The PARA page was originally published on 24 February 2023 and updated on 15 April 2026, so this is current guidance rather than an artefact.

**Distill.** This is the step people skip and it is the one the method hangs on. Progressive summarization is "summarizing and condensing a piece of information in small spurts, spread across time, in the course of other work, and only doing as much or as little as the information deserves." It runs in layers: Layer 1 is the raw capture, Layer 2 is "the first round of true summarization, in which I bold only the best parts of the passages I've imported", Layer 3 switches to highlighting "the 'best of the best,' only highlighting something if it is truly unique or valuable", Layer 4 is "an informal executive summary at the top of the note, restating the key points in my own words", and Layer 5 is a remix into something of your own.

Two guidelines come with it and both get dropped in the retellings. "More summarization is **not** better", because most notes should stop at Layer 1. And more highlights "dilute the discoverability" of a note, which is the failure mode of anyone who discovers the highlighter and applies it to everything.

**Express.** "turning incoming information into completed creative projects", on the principle that information is there to be put to use rather than consumed passively. The point of the first three steps is this one. A second brain that never produces anything is a reading list with extra steps.

So: capture on instinct, file by what you are actually working on, compress in passes, and ship something out of it. That is the method, it is sound, and you can start it this afternoon in any app that holds text.

## Why resonance only works for one person

Now the part the ranking pages do not get to.

Forte is candid about the risk in his own technique. "Summarization is risky", he writes, because "you may be making the wrong decision about what's important." For a personal brain that risk is bounded, because the person who made the wrong call is the person who will later notice and fix it. The bolding in Layer 2 is a message you left yourself, and you are fluent in your own shorthand.

Hand the same note to a colleague and every one of those layers stops carrying information. Bold text with no stated reason reads as emphasis without an argument. A highlight says "this mattered" without saying to whom or why. A Layer 4 summary written "in my own words" is written in one person's words, and the parts it leaves out are the parts that one person already knew. The notes are not badly written. They are correctly written for an audience of one.

This is why the pattern teams keep hitting is not a missing-notes problem. Everyone has notes. What is missing is a reason to trust someone else's distillation, and the method as published has no step that produces one. Capture by resonance needs an owner of the resonance. Distillation into "my own words" needs a reader who shares the vocabulary.

The fifth requirement, then, is not another verb to add to CODE. It is a constraint on the other four: **every note has to be legible to a named person who was not in the room.** In practice that means stating the decision and not just the evidence, dating it, and saying what you rejected, because the rejected option is the thing a newcomer will otherwise propose again in six months. None of that is a software feature. What software decides is whether the attempt is possible at all, which is the next two sections.

## Can the AI just be the second brain?

Three of the pages currently ranking for this query answer yes, and relocate the whole system into an assistant's memory. The appeal is obvious: capture becomes conversation and distillation becomes somebody else's problem.

The ceiling on that is published by the model vendors themselves. Anthropic's context window documentation describes the window as "a 'working memory' for the model", and is direct about what happens as it fills: "As token count grows, accuracy and recall degrade, a phenomenon known as context rot. This makes curating what's in context just as important as how much space is available." The same page describes compaction, which "automatically summarizes earlier parts of the conversation on the server, so the conversation can continue past the context window limit."

Read that against Forte and the two halves line up exactly. Compaction is Layer 2 to Layer 4 distillation, performed by a model, under a default prompt, at whatever moment the window happened to fill, with no record of what it dropped. Forte's warning applies with more force, not less: somebody is making the wrong decision about what is important, and this time nobody can see which decision it was. We wrote up the mechanics of this separately in [does the context window reset](/blog/does-context-window-reset), and the short version is that the window is rebuilt from nothing on every request, so nothing survives it that was not written down elsewhere.

An assistant is excellent at the Express step and genuinely useful at Distill when a person checks it. It is the wrong place for the repository, because "external, centralized, digital" was the whole definition and a context window is none of the three.

## What changes when the brain has a second reader

The method stays. What changes is where the files sit and who can open them, and this is where most team second brain plans quietly die.

The usual starting point is a folder of markdown in a sync service, or an Obsidian vault with Sync on it. That gets you further than people expect and then stops at a specific place, which Obsidian documents plainly. Its collaboration help page states that "Obsidian does not yet support collaborative live editing on the same file", that "You will not see the other user's cursor, and their edits will only appear once the changes are synced", that "The maximum number of collaborators on a shared vault is 20 users", that "All collaborators must have an active Sync subscription to access a shared vault", and that "All collaborators receive the same permissions as the vault owner, with one exception: only the vault owner can invite collaborators."

The last of those is the one that ends the plan. A second brain contains salary notes, a half-finished reorganisation, an unflattering account of a vendor. One permission level for everybody means either the vault is sanitised until it is useless or it is not shared.

Baalda's answer is to keep the files and put a server in the path that can refuse. Notes are plain markdown on your own disk, so the Capture and Distill steps are just editing text files and anything that reads text can read them. Several people edit the same note at once through a CRDT, which merges each keystroke rather than resolving a conflict afterwards, so two people distilling the same capture produce one note instead of two copies. Access is set per folder and per file, with each person's vault-level access as the ceiling, so Resources can be open to everyone while a single Projects folder is readable by three people. It is open source under Apache-2.0, free to run locally or self-host, and the managed backend is free for 2 people and 1 synced vault, with the Team plan at [$10 per seat a month or $110 per seat a year](/pricing), minimum 3 seats.

The AI half uses the same path rather than a second one. An assistant connects over MCP to the same files, with `read_note`, `search_notes`, `create_note` and `edit_note` for anchored replacements, and the permissions that apply to a person apply to it:

```bash
claude mcp add --transport http brain http://localhost:3010/api/mcp \
  --header "Authorization: Bearer mcp_…"
```

So a Layer 4 summary an assistant drafts lands as a visible edit in a file a person can correct, in a folder someone chose to grant it, which is the only arrangement where machine distillation is auditable at all.

## Where tooling stops helping

Baalda supplies none of the method, and this is the honest limit. It will not decide what resonates, it will not run the distillation passes, and it has no opinion about whether your folders match PARA. A shared vault with no method in it is the same pile with more authors and worse attribution. The teams that get value from this write the decisions down because someone insisted on it, and the tool only has to stop being the reason they cannot.

Two cases where it is the wrong pick. If you are one person and expect to stay one person, you do not need any of this: Obsidian plus the method as published is the better answer, and Baalda adds a server for no benefit. And if what you actually want is Notion's relational databases, rollups and views, markdown files will not get you there, no matter how well they sync. Our [round-up of second brain apps](/blog/best-second-brain-apps-2026) is organised around which of those trades you are making, and the [team second brain](/blog/team-second-brain) piece is the longer version of the argument above.

The method is forty minutes of reading and the rest is habit. The only structural decision is the one in the definition: external, centralized, and digital, for how many readers.

## FAQ

### What is a second brain?

Tiago Forte, who popularised the term, defines it as "an external, centralized, digital repository for the things you learn and the resources from which they come". External so it is not in your head, centralized so there is one of it, and digital so it can be searched rather than remembered. The definition says nothing about how many people read it, which is the part that decides whether a team's version works.

### What is the PARA method?

PARA is the organising half of the method: four folders, sorted by how active the material is. Forte defines Projects as "short-term efforts ... that you take on with a certain goal in mind", Areas as "important parts of your work and life that require ongoing attention", Resources as "a range of topics you're interested in and learning about", and Archives as anything from the other three that is no longer active. The page was updated on 15 April 2026.

### How long does it take to build a second brain?

The reading is one afternoon and the structure is four folders, so the setup is not the work. Forte's own guidance points the other way: most notes should stop at Layer 1 of distillation, because "More summarization is not better". The habit of capturing and distilling is what takes months, and it is the part no app installs for you.

### Can an AI be my second brain?

It is a good place to do the work and a poor place to keep it. Anthropic's documentation calls the context window "a 'working memory' for the model" and states that "As token count grows, accuracy and recall degrade, a phenomenon known as context rot", with compaction summarising older turns on the server to continue past the limit. A repository that silently rewrites itself is not external, centralized or digital in the sense the definition means.

### Which app should I build a second brain in?

Alone, Obsidian plus the method as published, and you do not need anything else. For a team the constraint is permissions: Obsidian's help page states that "All collaborators receive the same permissions as the vault owner" and that it "does not yet support collaborative live editing on the same file". That is the point where a tool with per-file access and real co-editing, such as Baalda, starts to matter.