Building a Second Brain with AI: My System for Never Forgetting Anything

The Problem With Your Brain

Your brain is magnificent at having ideas. It is terrible at keeping them.

I know this because I spent years losing thoughts. Not small thoughts — big ones. The kind of insight that hits you in the shower at 7 AM and is completely gone by 7:15. The article you read last month that was exactly relevant to what you're working on now, but you can't remember where you read it or what it said or even what it was about, just that it existed and it was good.

This is the fundamental knowledge management problem: the gap between what you encounter and what you can retrieve when you need it. And it's a problem that, for most of human history, we've solved poorly. Notebooks that fill up and disappear. Bookmark folders with 3,000 unsorted links. Note-taking apps that become digital junk drawers.

I tried them all. Evernote, Notion, Obsidian, Roam, Apple Notes, physical notebooks (three different systems), index cards, and at one desperate point, a voice recorder I carried around like a journalist from the 1970s.

Nothing stuck. Not because the tools were bad, but because every system required me to do something my brain resists: organize in the moment of capture. When I have an idea, I want to capture it. I do not want to decide what folder it goes in, what tags it gets, or how it relates to my other ideas. That's a second task layered on top of the first, and it creates just enough friction to guarantee I'll stop doing it within two weeks.

Then I found a different approach. One that uses AI to handle the part I'm bad at — organization, connection, retrieval — while I focus on the part I'm good at: having thoughts.

The Architecture (Simple, I Promise)

My second brain has three components:

1. A capture layer — where everything goes in, with zero friction
2. An AI processing layer — where things get organized, tagged, and connected automatically
3. A retrieval layer — where I get things back when I need them

That's it. Three layers. Let me walk you through each one.

The Capture Layer

I capture everything into plain text files using the FFilesystem MCP. This might sound old-fashioned, and it is. Deliberately so.

Plain text files are:
- Universally readable (no proprietary format lock-in)
- Incredibly fast to create
- Searchable by any tool
- Nearly impossible to corrupt
- Compatible with everything forever

My capture process is simple: when I have a thought, encounter something interesting, or want to save information, I dump it into a text file. No formatting required. No organization required. No metadata required.

The file naming convention is equally simple: date and a brief description. 2026-04-08-interesting-article-about-memory.txt. That's it. Everything goes into a single /notes/inbox/ directory.

This is the crucial insight: the capture layer should have zero friction. If you have to think about where something goes, you've already introduced enough resistance to kill the habit. Everything goes in the inbox. Everything. Every time. No exceptions.

I capture probably 15-20 items per day. Some are full paragraphs. Some are single sentences. Some are links with no commentary. Some are half-formed ideas that look like nonsense to anyone but me. It doesn't matter. Into the inbox they go.

The AI Processing Layer

Once a day — I do this in the evening, but the timing doesn't matter — I run my notes through an AI processing step. This is where the magic happens.

Using CContext7 for looking up any reference documentation I need, and the FFilesystem MCP for the actual file operations, I have a workflow that:

1. Reads all new inbox files. Everything captured that day.

1. Categorizes each item. Not into rigid folders, but into flexible categories based on content analysis. The AI reads each note and determines what it's about — is it a project idea? A reference? A personal reflection? A piece of information I might need later?

1. Generates tags. Not the tags I would assign (which would be arbitrary and inconsistent), but tags based on actual content analysis. The AI identifies themes, topics, people mentioned, projects referenced, and emotional tones.

1. Creates connections. This is the most valuable step. The AI compares each new item against my existing notes and identifies potential connections. "This thought about productivity systems is related to your note from March 15th about cognitive load." "This article about nutrition connects to your ongoing research into morning routines."

1. Generates a daily summary. A brief digest of what I captured, what themes emerged, and what connections were found. This summary becomes a note itself, creating a searchable log of my intellectual activity over time.

1. Moves processed files from the inbox to an organized archive, with the AI-generated metadata embedded in each file.

The whole process takes about 5 minutes. And it does the thing I could never do: organize consistently, comprehensively, and without the cognitive overhead that made every previous system fail.

The Retrieval Layer

This is where the second brain pays dividends.

When I need to find something — a specific note, a cluster of related ideas, everything I've captured about a particular topic — I query my system using natural language.

Not keyword search. Natural language. "What were my thoughts about time management from the last month?" "Find everything related to the research project I started in February." "What connections exist between my notes on AI and my notes on education?"

This is fundamentally different from traditional search. Keyword search requires you to remember the exact words you used. Natural language retrieval works with concepts, themes, and intentions. The AI understands that a note about "feeling overwhelmed by too many projects" is related to a note about "prioritization frameworks" even though they share no keywords.

The retrieval layer has transformed how I work. Before, I was constantly recreating ideas I'd already had because I couldn't find the original. Now, nearly every idea I have connects to something I've captured before, building a web of knowledge that compounds over time.

My Actual Daily Workflow

Let me be concrete about what this looks like in practice.

Morning (5 minutes):
I check yesterday's daily summary. This serves two purposes: it reminds me what I was thinking about yesterday, and it often sparks new connections that I capture immediately (which go into today's inbox).

Throughout the day (30 seconds per capture):
Whenever I have a thought, read something interesting, or want to save information, I drop it into the inbox. Quick, no friction, no organization.

Evening (5 minutes):
I run the AI processing step. Review the connections it found. Sometimes I'll expand on a note that the AI flagged as potentially important, or merge two notes that the AI identified as related.

Weekly (15 minutes):
I do a broader review — looking at themes from the week, identifying ideas that are growing into potential projects, and cleaning up any misclassifications from the AI processing.

Total daily time: about 10 minutes, plus however long the individual captures take (which is genuinely about 30 seconds each).

What Makes This Different From [Every Other System]

I've tried enough knowledge management systems to write a book about them. Here's what makes this approach actually stick:

1. Capture and organization are completely separated. I never have to think about where something goes in the moment I'm capturing it. This is the fundamental design principle, and every other system I've tried violates it. Notion wants you to pick a database. Obsidian wants you to think about links. Even a physical notebook requires you to decide which notebook. Here, everything goes in one place, and the AI sorts it later.

2. The AI handles the boring work I'll never do. Tagging, categorizing, finding connections — these are tasks that I know I should do and will never do consistently. The AI does them without complaint, without inconsistency, and without taking a day off.

3. Natural language retrieval eliminates the "what did I call it" problem. With keyword search, you need to remember your own terminology. With natural language retrieval, you just describe what you're looking for. This is an enormous quality-of-life improvement that I genuinely cannot overstate.

4. It gets better over time. The more notes I add, the more connections the AI can find. The system has genuine compound interest — each new idea is enriched by every previous idea. After a few months, the web of connections becomes genuinely surprising and useful.

5. It's tool-agnostic. Because the foundation is plain text files, I'm not locked into any particular app or service. If I want to switch my AI processing layer, I can. If I want to add a visual interface, I can. The data is mine, in the simplest possible format.

Real Examples of the System Working

Example 1: The Accidental Connection
Three months into using this system, I captured a note about a podcast episode discussing decision fatigue. A week later, I captured a note about a friend's complaint about being overwhelmed by choices at the grocery store. The AI processing step connected these two notes and also linked them to an older note I'd written about simplifying my wardrobe.

Result: I wrote an article about "choice architecture for everyday life" that connected ideas from three different contexts. I never would have made those connections manually.

Example 2: The Research Accelerator
When I started researching a new project, I queried my second brain for everything related to the topic. It returned notes from the last eight months — articles I'd read, ideas I'd had, conversations I'd flagged — organized thematically. What would have been a "starting from scratch" research phase became a "building on existing knowledge" phase. I had a comprehensive brief before I wrote a single word.

Example 3: The Problem-Solver
I was stuck on a design challenge and, on a whim, asked my system: "What approaches to simplification have I captured?" It returned notes from completely unrelated contexts — a note about Japanese design philosophy, a note about how a particular chef describes recipe development, a note about code refactoring — and the combination sparked an approach I wouldn't have arrived at through deliberate brainstorming.

The Honest Downsides

Setup takes effort. Getting the pipeline running — the FFilesystem MCP configured, the AI processing workflow built, the retrieval system tuned — took me a weekend. It's not plug-and-play (yet).

The AI isn't perfect at categorization. Maybe 85-90% of the time, the AI categories and connections are useful. The other 10-15%, they're wrong or irrelevant. This is why I do a brief review each evening rather than fully trusting the automation.

You need to actually capture things. The system only works if you feed it. If you have a thought and don't capture it, it doesn't exist in the system. This requires developing a capture habit, which takes about two weeks to become automatic.

It can feel overwhelming. After six months, I have thousands of notes with tens of thousands of connections. Sometimes querying the system returns more information than I want to process. I'm still developing strategies for managing information abundance (which is a much better problem than information loss, but a problem nonetheless).

Getting Started (If You Want To Try This)

1. Create an inbox folder. Anywhere. Desktop, Documents, wherever you'll actually use it.

1. For one week, just capture. Don't organize. Don't tag. Don't think about it. Just dump thoughts, links, and ideas into text files in that folder.

1. At the end of the week, try using an AI to categorize and connect your notes. You can use CContext7 to look up any reference materials you need, and even a simple AI chat can identify themes and connections in your captured notes.

1. See if the connections surprise you. If they do, you've found the value. If they don't, capture for another week. The system needs density to produce interesting connections.

1. Scale from there. Add the Filesystem MCP for smoother file operations. Build the automated processing pipeline. Develop your retrieval queries.

Start small. The system is only as good as your capture habit, and habits are built one day at a time.

A Final Thought on Memory and Meaning

Here's something I didn't expect: building a second brain changed how I experience my first brain.

Knowing that I have a reliable capture system freed my biological brain from trying to hold onto everything. I became less anxious about forgetting things. I became more present in conversations, because I wasn't frantically trying to remember key points — I'd capture them later. I started having more ideas, because my brain wasn't clogged with trying to retain the old ones.

My second brain doesn't think for me. It remembers for me. And that distinction — between thinking and remembering — turns out to be more important than I realized.

Your brain wants to create, connect, and imagine. Let it do that. Let something else handle the filing.