What is a second brain?
Hey, it's Ilan's first brain here. A second brain is basically a system for keeping track of your life so you don't have to hold everything in your head. Mine puts information in the tools that handle it best, then uses one AI assistant to help me find things, connect the dots, and get things done. Here's how mine works and how you can build yours.
1. Map your life as it works today
Start by mapping two things: the information you keep and the routines you use to manage it. Where do your notes, journal, calendar, tasks, files, and lists live today? How do you capture, review, and act on them?
You do not need to digitize everything. A handwritten journal can stay outside the system. I now dictate my journal entries because the text is easy for me to capture and useful to my assistant, but that is a choice, not a requirement.
2. Choose systems that work for you and your AI
Change only what is worth improving. Each system should work well for you directly and be accessible to your AI when that is useful. Together, they become the small databases of your life. Here is my setup.
The Plugin badge means the tool connects directly through ChatGPT's plugin directory.
| System | What it owns |
|---|---|
| Plugin Tasks, follow-ups, and waiting loops | |
| Plugin Meetings, appointments, and travel | |
| Plugin Email threads and communication state | |
| Plugin Connects health records and apps; providers remain the source of truth | |
| Plugin Connects accounts through Plaid; institutions remain the source of truth | |
| Plugin Journaling, raw thoughts, personal context, and a thin system map | |
| Plugin Code, architecture, decisions, and working context | |
| Local files, active projects, and private material | |
| Books and reading history | |
| Films, watchlists, and viewing history | |
| Saved places | |
| Sick achievements |
Why Todoist: I tried Apple Notes, Reminders, Notion, Trello, and others before settling on Todoist. It gives me projects, fast capture, and tasks alongside my calendar, while ChatGPT can read and update the same live state. I can add or finish a task myself, and the next conversation starts with the right information.
Why not one repository: I originally tried to keep the whole Second Brain in one GitHub repository. Now I let purpose-built systems own the data they handle best, including permissions and sharing, while ChatGPT provides the wider view across them. The personal repository is one component, best for journaling, raw thoughts, personal context, and a thin system map.
3. Set up one AI interface
Choose one primary AI assistant to work across the system. I use ChatGPT because I like the interface and its growing library of native plugins. That may change. Claude also works. I have not used Gemini enough to judge it.
| Layer | Set it up with |
|---|---|
| Instructions | System map, communication style, and approval rules |
| Memory | Durable preferences that should follow you between conversations |
| Skills | Narrow, repeatable procedures with a clear trigger and output |
| Plugins | Connections that provide access to your data and other tools |
| Scheduled tasks | Workflows that genuinely need a clock |
Skills
Think of a skill as the playbook you would give the world's best intern. Explain the job once, including your preferences, steps, edge cases, and desired output, then improve it as you learn. Your assistant can load that playbook whenever the same kind of request comes up, so triggering the workflow is enough and you do not have to repeat all its little nuances.
Here are a few skills I use:
| Skill | What it does |
|---|---|
| Morning brief | Combines my calendar, tasks, relevant email, and preferred AI news into one prioritized daily view |
| Find flights | Finds and compares flights using my preferred airports, airlines, timing, baggage rules, and format |
One annoying limitation: ChatGPT and Codex currently maintain separate skill libraries. I keep my personal skills in ChatGPT and add local Codex skills only when a workflow needs access to my computer. Ideally, I could choose one source of skills and use it across both, but they do not sync automatically today.
4. Use the system
Once the system is connected, the value becomes less abstract. Here are a few things I've had it do:
| Example | What happened |
|---|---|
| Bike rebate | Researched the rebate, found the right contacts, drafted the emails, timed the follow-ups, and managed the thread until the rebate was approved |
| Meeting follow-up | ChatGPT turns meetings into summaries, follow-up drafts, and tracked action items, then asks for approval before acting |
| Working remotely | With my Mac plugged in at home, ChatGPT can do essentially everything I can do on my Mac, even while I'm on the go |
Operating rules
- One source of truth. Keep each fact in one system.
- Use purpose-built databases. Choose the best small database for each kind of information, such as Google Maps for places or Todoist for tasks.
- Go to the source. For changing information, have the AI check the system that owns it instead of relying on a copied version.
- Keep the system simple. Start lean, prefer low-friction workflows, and let complexity earn its place.
- Keep iterating. Test your workflows, correct misses, and refine the system over time.
- Put procedures near their scope. Keep general procedures in Skills and project procedures in project repositories.
- Start with a short leash. Keep a human in the loop for consequential actions. Expand the AI's autonomy only as each workflow earns your trust.
- Protect sensitive information. Keep it local unless you explicitly decide otherwise.
- Understand the architecture. Know which systems own what, how information moves between them, and what the AI can change. Convenience should not make your own system opaque.
- Be articulate. Say what outcome you want, where it should go, and which constraints matter. Vague requests usually produce vague results.
- Have fun with it. Build something you actually enjoy using.