How to Build an AI Second Brain

Updated January 2026 | 9 min read

A second brain captures what you know so you don't have to remember everything. For humans, that means note systems. For AI, it means giving your assistant access to files that contain your context.

This is a 90-minute setup process. You'll create a folder structure, write one master file that AI reads every session, and organize your existing information so AI can find it.

The Core Concept

AI starts every conversation with amnesia. It doesn't remember yesterday's session. It doesn't know your business, your preferences, your processes, or your projects.

A second brain for AI solves this by storing your context in markdown files that AI reads at the start of each session. One master file (CLAUDE.md) tells AI who you are, what you do, and where to find specialized information.

When you ask AI to help with a client project, it reads CLAUDE.md, sees you have a Clients folder, and knows where to look for that client's history. You don't explain. AI reads.

The 90-Minute Setup

Set a timer. This is not a research project. You're building scaffolding, not a cathedral.

Minutes 0-15: Create the folder structure. Make four top-level folders: Projects, Processes, Clients (if applicable), and Reference. These hold different types of information.

Projects: active work with deadlines. Processes: repeatable workflows. Clients: client-specific context. Reference: background information AI might need.

Minutes 15-45: Write CLAUDE.md. This is your master file. It lives in the root directory. AI reads it first every session.

Start with identity: your name, role, company. What you do, who you work with, what your responsibilities are.

Add your schedule. When you're available, when you're in meetings, what your typical day looks like. AI can't schedule around conflicts it doesn't know about.

List your projects. One line per project. Include status: active, paused, waiting. AI needs to know what's in motion.

Document your tools. What CRM you use, what email system, what project management software. If AI will interact with these systems, write down the names.

Add your preferences. How you like reports formatted, what tone to use in client emails, whether you want data tables or summaries. These aren't rules—they're defaults AI can reference.

Minutes 45-75: Populate the folders. Go through your existing documents, notes, emails, and bookmarks. Move relevant files into the folder structure.

Don't rewrite anything. Just move it. If you have a process doc in Google Docs, export it as markdown and drop it in the Processes folder.

If you have client notes in email, copy them into a markdown file in the Clients folder. One file per client. Name it with the client name.

If you have project specs scattered across Slack, Notion, and your desktop, gather them into Project folders. One folder per project.

Minutes 75-90: Test it. Open Claude Code (or Claude with Obsidian vault access). Point it at your folder. Ask it a question that requires context from your files.

"What's the status of the [project name] project?" or "What's our process for onboarding new clients?" or "Pull up the notes for [client name]."

If AI answers correctly without you explaining, your second brain works. If it asks clarifying questions, note what information it needed and add it to the relevant file.

File Naming That AI Understands

AI searches by filename before it reads content. Descriptive names mean faster retrieval.

Use kebab-case (lowercase with hyphens): "client-acme-corp.md" not "Client Acme Corp.md" or "clientAcmeCorp.md". Consistent formatting means consistent search results.

Start with category: "project-website-redesign.md", "process-invoice-approval.md", "client-north-industries.md". This groups related files in directory listings.

Avoid generic terms. "Notes.md" tells AI nothing. "meeting-notes-2026-01-15.md" tells AI this is a specific meeting from a specific date.

Include dates when relevant. "report-q4-2025.md" or "contract-signed-2026-01-10.md". Dates provide temporal context.

What to Include in Context Files

Each domain (Projects, Clients, Processes) needs different information.

Project files: Goal, deadline, stakeholders, status, current blockers, next actions, deliverables, links to related files.

Client files: Company name, contact names and roles, contract terms, communication preferences, project history, open issues, renewal dates.

Process files: When to use this process, step-by-step instructions, decision points, required tools, expected timeline, owner.

Reference files: Product specs, brand guidelines, code snippets, templates, research, credentials (except passwords—never store passwords in plaintext).

Each file should answer the question: "What does AI need to know to help me with this?"

Maintaining Your Second Brain

Add information when you notice AI doesn't have it. If you explain something to AI and realize it should have known this already, write it down immediately.

Update project statuses when they change. When a project moves from active to paused, update the file. AI can't help you prioritize if it thinks everything is active.

Archive completed work. Create an Archive folder. Move finished projects there. This keeps your active workspace focused.

Review CLAUDE.md monthly. Your role changes, your projects change, your preferences change. The master file should reflect current reality, not last quarter's reality.

Don't let perfect be the enemy of useful. A second brain with 80% of your context is infinitely better than explaining everything from scratch every session.

When It Clicks

You'll know your AI second brain is working when you ask AI to do something and it just does it—no follow-up questions, no clarifications, no "Can you provide more context?"

You'll stop prefacing requests with background information. You'll stop saying "As I mentioned last time..." because AI already knows.

Your sessions become execution, not explanation.