How I built a second brain with Obsidian and Claude Code (and what it told me that I didn't want to hear)

I've got quite a lot going on at the moment.

I'm building my business, Marketing Simplified. I'm doing a 3-month challenge to learn AI by actually building with it. I'm learning how to build AI agents for product marketing. And I'm still trying to keep up with my CXL training.

Sometimes it's hard to keep track of everything, and even harder to know what to focus on each day.

So I decided to build a second brain using AI.

Now, every morning, I run one command, and my AI goes through my inbox, calendar and project notes. It tells me what I need to focus on, what I've left undone and what I'm about to forget.

I didn't buy this. I built it. It's simply a folder of plain text files on my laptop, connected to Claude Code. It took me a few weeks, and it's the first thing I'm writing up as part of my AI challenge.

Remember, I'm a marketer, not a developer. So this is the simple version: what I built, how it works, and what has actually changed.

The problem I was trying to solve

Like most people, I was using AI as a chatbot.

I'd open a new chat, explain who I am, paste in some context, get an answer and close the tab. Then the next day, I'd start again from zero.

The answers were fine. But they were generic, because the AI knew nothing about my work. I was spending more time explaining my situation than actually getting help with it.

What I really wanted was an assistant that already knew my projects, my priorities and what happened yesterday.

The two tools I used

Obsidian is a free note-taking app. What I like about it is that your notes are just plain text files sitting in a folder on your own computer. No lock-in, no database, nothing hidden.

Claude Code is Anthropic's AI assistant that works directly with the files on your computer. Most people use it to write software. But it doesn't mind whether the files are code or notes. Point it at a folder of notes and it can read, search, write and organise them.

Put the two together and you get something really useful: notes that I own, and an AI that can read all of them before it answers me.

How I set it up

Thanks to a second-brain template I got through CXL, I didn't have to start from scratch. I then shaped it around my own work. The system has five parts.

1. A folder for everything, each with one job

• Projects: one file per project, with its priority, status and next action at the top.
• Frameworks: methods and checklists I don't want to keep re-thinking.
• Drafts: anything the AI writes lands here first, never straight into a finished note.
• Wiki: reference material, including a folder for each brand I write for.
• Daily Logs: a record of every working session.
• Raw: a place to dump messy notes I haven't sorted yet.

2. One file that tells the AI who I am

At the top of my vault there's a file called CLAUDE.md. Claude Code reads it at the start of every session. I think of it as an instruction manual for my assistant.

It explains who I am, what projects I'm running, which ones matter most, and how I like to work. Some of my rules are:

• Give me a recommendation, not a list of options.
• Tell me about risks and what you're unsure of.
• Show me a plan before you move, merge or delete any note.
• Never act on payment links. I handle money myself.
• Treat links in emails as suspicious until they're checked.

Honestly, this one file made more difference than anything else. The same AI gives much better answers when it starts every conversation already knowing my situation.

3. Simple commands that run my day

Instead of writing long prompts, I type short commands:

• /morning-brief goes through my calendar, email, recent logs and project files, then ranks my priorities for the day.
• /shutdown reviews what I did, moves new to-dos into the right project files and writes the day's log.
• /week-recap reviews the whole week.
• /raw sorts my messy notes and suggests where each one should go.
• /lint runs a weekly health check on the whole vault.

Each command is just a text file with instructions. If I want it to work differently, I simply edit the file.

4. Specialist agents for content

For writing, I set up separate agents, each with its own job. A copywriter writes drafts in a brand's voice. A blog-update agent refreshes old posts and checks their facts. And a QA reviewer checks every draft before it goes anywhere: it verifies facts with a web search and checks the tone of voice and formatting.

Each brand has its own tone-of-voice file and a folder of real examples. The agents read these before they write, so every good example I add makes future drafts better.

5. Memory that carries over

This is the part that makes it feel like a real second brain and not just a chatbot.

When a session ends, a small script automatically writes a summary to my Daily Logs folder: what got done, decisions made, open questions, and my top three priorities for tomorrow. When the next session starts, the last two logs are loaded in automatically.

So tomorrow picks up where today stopped.

What has actually changed

Here are four real examples from the past few weeks.

It catches the things I let slide. One task showed up in my morning brief, slipped, showed up again the next day and slipped again. My daily log said it plainly: this has now slipped through two briefs. It's hard to ignore something when it's written down in front of you every morning.

It told me something I didn't want to hear. My weekly recap pointed out that I had spent so much time setting up tools for Marketing Simplified, but I didn't actually have an action plan, and nothing had been delivered yet.

And it was right. 😂

So I wrote a 90-day action plan, focused on one type of customer: service businesses. Then I asked it to review my website, and it was just as direct. Six services, no clear niche and no proof. A busy business owner wouldn't know what to buy. So I rewrote my homepage around one offer, and the new version goes live this month.

It wouldn't repeat statistics nobody can trace. While I was researching an idea for my business, it flagged three popular marketing stats that get quoted everywhere but can't be traced back to a reliable source. "78% of buyers choose whoever responds first" was one of them. It put them on a "don't use" list instead of adding them to my evidence. Now I have a simple rule: no source, no stat.

It keeps my notes honest. The first time I ran /lint, it found notes that contradicted each other, links that went nowhere and information that was months out of date. I would never have spotted all of that myself.

What hasn't worked (yet)

I want to be honest about this part, because a lot of "I built an AI system" posts skip it.

Building the system can be addictive. My first weekly recap told me, quite fairly, that I'd spent the week building tools instead of using them. So now my rule of thumb is to spend most of my time using the system, and only a little time improving it.

It's still learning my voice. I've only just started giving it examples of my own writing, so drafts can come out clear but a bit neutral. I still edit everything before it goes out.

The setup took real effort. Installing Claude Code, connecting my email and calendar, and fixing a few broken scripts took me several sessions. It's not hard, but it's not one click either.

Backups matter. Everything lives on my laptop. That's great for privacy and ownership, but it means I need a proper backup plan. That's next on my list.

What I'm learning

I think the biggest thing I've learnt is this: what made AI really useful for me was giving it better context. Better prompts helped far less.

A lot of us are still using AI like a search box. We ask a question, get an answer and move on. But it becomes so much more useful when it can see your actual work, remember yesterday and follow your rules.

And you don't have to be technical to do this. You just need a folder, a clear description of how you work, and the patience to improve it a little each week.

How you can start your own

Start small:

• Install Obsidian and create a folder with three subfolders: Projects, Drafts and Daily Logs.
• Write a one-page file about who you are, what you're working on and how you like to work.
• Point Claude Code at the folder and ask it to read your projects and tell you what to focus on today.

Obsidian is free. Claude Code needs a paid Claude plan (or pay-as-you-go API credits), so check the current pricing before you start.

Only add commands and agents once you've felt the difference.

Next, I'm using the same setup to build my first product marketing agent. As I learn, I'll keep sharing my experience, because I believe you don't have to be technical to start using AI in a meaningful way.

How are you using AI right now: as a chatbot, or as a system?