Build Your AI Second Brain

Every frontier AI tool builds context about you over time — but that memory is locked inside each one. Switch from ChatGPT to Claude and you lose everything the previous tool knew about you. You end up judging the new model against the old harness — which isn't a fair comparison. A second brain inverts this: you build the memory once, you own it, and you bring it to whatever tool you're working in.

Models get the headlines. Harnesses deliver the results. Dharmesh's analogy: the AI model is the engine. The harness — the system you build around it — is the car. An engine alone doesn't get you anywhere. A second brain is the harness that makes every model you use dramatically more useful.

The value of a second brain isn't immediate — it's cumulative. Every connection you add, every fact the system learns about you, every briefing it gets right makes the next one better. Dharmesh describes getting a daily executive briefing email that has gotten progressively smarter as his second brain has grown. He replies to it with corrections and preferences, and those become facts the system carries forward into every future briefing.

2025 was a record year for new company formations in the US — nearly 5 million new businesses. 86% are not expected to hire a single employee. AI is making it possible for one person to do what previously required a team. A second brain is the infrastructure that makes that possible at scale.

Outcome: Clean vault with LLM-ready folder architecture, configured in under 15 minutes

Download Obsidian at obsidian.md. Create a new vault. Name it something functional like “Marketing-Brain.” Point it at a local folder you can find easily, or a Dropbox/iCloud folder if you want automatic sync across devices.

Two settings that matter: set default editing mode to Source mode (keeps your markdown clean when pasting into AI context windows), and set new notes to default to your Inbox folder. Everything lands there first, gets sorted later.

Then build this folder structure:

Marketing-Brain/

├── 00-Inbox/

├── 01-ICP-and-Positioning/

├── 02-Campaign-Learnings/

├── 03-Competitor-Intelligence/

├── 04-Content-and-SEO/

├── 05-Frameworks-and-Mental-Models/

├── 06-Research-and-Data/

├── 07-AI-Prompts-and-Workflows/

├── 08-Weekly-Reviews/

└── _Templates/

How to create the folders

In Obsidian's left sidebar, right-click anywhere in the file explorer and select New folder. Type each folder name exactly as shown and hit enter. Two minutes total.

Once all ten folders exist: Settings → Files and Links → Default location for new notes → In the folder specified below → type 00-Inbox.

The numbering forces consistent sort order so navigation is always fast. The underscore prefix on _Templates sorts it to the bottom — it's infrastructure, not content.

Outcome: Three foundational files that transform every AI conversation you have

These files aren't personal notes. They're structured briefing documents written to be read by an LLM. Use clear headers, write in declarative statements, keep each file focused on one topic.

File 1 — ICP Master (01-ICP-and-Positioning/ICP-Master.md)

• Who your buyer is
• What they're afraid of
  How they talk about the problem
• What they've already tried
• Where they find solutions
• Positioning statement
• Key differentiators
• Messaging landmines (things to avoid saying, language that signals wrong fit, etc.)

File 2 — Campaign Learning Template (_Templates/Campaign-Learning-Template.md)

Every campaign that concludes gets a file from this template:

• What was the hypothesis
• What were the results
• What specifically worked (not "the creative was good" but "the hook led with a counterintuitive stat and beat our control by 34%")
• The transferable principle.

You’re building a proprietary database of what actually works for your specific audience.

File 3 — My AI Context Block (07-AI-Prompts-and-Workflows/My-AI-Context-Block.md)

300–500 words. Who you are, what you're building, your audience, how you think about marketing, your current focus, your voice and tone.

Paste this at the top of every AI conversation before you start. Update it monthly. This is the single most leveraged document in your vault — start here if you do nothing else.

Step 2 complete when:

• ICP-Master.md exists with real content (not placeholder text)
• Campaign-Learning-Template.md is in _Templates/
• My-AI-Context-Block.md has been used in one real AI work session

Outcome: A knowledge graph where related ideas surface together — automatically

Cross-linking is what separates a folder of markdown files from an actual knowledge graph. In Obsidian, you link using [[double brackets]]. When you reference [[ICP-Master]] inside a campaign learning note, Obsidian tracks that relationship bidirectionally — every file that references your ICP shows up in the ICP file's backlinks automatically.

Create a Map of Content (MOC) file for each major folder — an index that links to everything in that section. When working with AI, paste the MOC first. Your AI sees the full landscape of what you know about a topic, then you decide together which specific files to load based on the task.

Add frontmatter tags to every file — status (current/outdated) and type (reference/capture). Takes 20 seconds per file. Means you can quickly identify what's worth loading before you paste anything.

Outcome: Any article, thread, or YouTube transcript in your vault in under 60 seconds

If it takes more than 60 seconds to get something into your vault, you might not do it consistently.

Install MarkDownload from the Chrome Web Store (free, open-source). Configure it to point at your 00-Inbox folder. Set your filename template to {date:YYYY-MM-DD}-{title} so every captured file is timestamped and scannable. Add a frontmatter template that auto-tags everything as "unprocessed" on capture.

Now when you're reading anything worth keeping, you hit one button. It lands in your inbox with a full transcript, properly named and tagged.

Outcome: Natural language queries across your entire knowledge base in under 2 minutes

You don't need to write code — you need to know how to use a code tool as a natural language interface for working with your vault at scale.

Your options:

• Cursor — Best for non-developers. Visual interface with AI chat side-by-side. Open your vault folder in Cursor, ask questions across all your files. Recommended starting point.
• Claude Code — Best if you're already in the Claude ecosystem. Conversational file operations in plain English. Steeper setup (terminal-based).
• Codex (OpenAI) — What Matt uses. Similar to Claude Code but with slightly more UI structure. Available via your existing ChatGPT account.

Open your vault folder in your chosen tool. You can now ask: "Read all files in 02-Campaign-Learnings and give me the three most consistent patterns you see." That operation takes 30–60 minutes manually. With this setup: under 2 minutes.

Create a Vault-Query-Prompts.md file in 07-AI-Prompts-and-Workflows/ and start populating it with your most useful queries:

• "Read [folder] and identify the 3 most consistent patterns or learnings"
• "I'm about to work on [project]. Which 3–5 files from my vault would be most relevant?"
• "What are the biggest gaps in my ICP knowledge based on what's in 01-ICP-and-Positioning?"
• "Find every note that mentions [competitor] and summarize what we know about them."
• "Compare my last five campaign learnings — what principle keeps showing up?"

Outcome: A 30-minute weekly process that keeps the system compounding without turning into a second job

Block 30–45 minutes weekly. Put it on your calendar as a recurring event. The vault without a review process is just a fancy bookmarks folder.

Part 1 — Process the Inbox (10 min)

For each file in 00-Inbox, make one of three calls:

• Promote — Move to the right folder, update frontmatter to current, add wiki-links, remove inbox tag.
• Extract and archive — The source isn't worth keeping, but there's one insight in it. Create a new note with just that insight, properly tagged. Delete the raw capture.
• Delete — You captured it in the moment but it doesn't add to your vault. Be ruthless. A smaller, sharper vault beats a comprehensive but noisy one.

Part 2 — Update one core file (10 min)

Pick one file from your core documents — ICP Master, a competitor file, your AI context block — and spend 10 minutes updating it. Add new information, revise anything no longer accurate, add new links. One file per week. That's 50+ substantive updates to your core knowledge per year.

Part 3 — Create one new note (5–10 min)

A framework, a campaign learning, something you've been thinking about this week that isn't in your vault yet. Use a template. Keep it tight.

Part 4 — Identify your AI work for the week (5 min)

Which 2–3 vault files would give you the most leverage in your work this week? Write them down. This forces you to actually use the vault, not just tend it.

Monthly: Run a synthesis query across a vault section. "What do I now know that I didn't know 90 days ago?" The output becomes a new note in 08-Weekly-Reviews/. Over time, these become your most valuable documents.

Quarterly: Rewrite My-AI-Context-Block.md from scratch. Don't edit the old version — start fresh. Archive the old one with a date stamp. Looking back at previous context blocks 12 months from now will show you how your thinking has evolved.

Most AI companies — and most SaaS companies — don't publish RSS feeds. So Matt built an automation that watches their website sitemaps instead, alerting him the moment a new page is added. He knows about competitor moves within minutes of them going live. The same logic applies to your competitive landscape: new product pages, pricing changes, job listings that signal strategic direction. Build the agent once, get the intelligence forever. Start with your top three competitors and your top two AI tools of choice.

Once your vault has real content in it — campaign learnings, ICP notes, competitor files — start brain-dumping your day directly into Codex, Claude Code, or Cursor. Current projects, open problems, ideas you haven't figured out yet. The system cross-references your journal against everything you've saved and surfaces relevant content you'd forgotten you had. The output stops being generic AI advice and starts being advice grounded in your specific experience. This is the highest-leverage daily habit in the entire system — but it only works once the vault has enough content to draw from.