Claude, you're fired!
I just moved my fiancée off Claude Code (@claudeai) and onto @bot. We canceled her Claude Code subscription.
That's a first for me. Since February, whenever someone asked which agent to run their second brain on, I always said Claude Code. This is the first time since second brains launched in January I've pointed someone at something else.
I see a lot of people on X having their "agent moment" right now.
They're realizing an agent can start to run their computer, their business, and even their life, all from their phone. Claude just dropped Duet yesterday, so a lot more people are about to have theirs.
Mine was in February.
We've been pushing hard on AI in our personal lives and businesses ever since, which checks out for someone who's been building automation software for more than a decade.
Over the following months, I built my own second brain, and it's now over 35,000 files. More than 90% of my work runs through an agentic operating system (aka my second brain). At my company, it's more than half of the work (aka the company brain).
I even set up a second brain for my fiancée, who is also a founder - we've been toxenmaxxing ever since.
Which is why this move surprised me.
What is an AI second brain?
Before I explain why I moved her brain, you might be asking what a second brain is in the first place... A second brain - specifically an "AI-powered second brain" - is an extension of your own brain. It holds everything you've collected about your life, all in one place an agent can read: notes, plans, conversations, trips, photos, health, money, projects, and the half-formed ideas you jotted on your phone.
Then you ask it anything. What was the name of that wine we loved at dinner a few years ago? When did the car last get new tires? What did I give my mom for her birthday two years ago? What was the winning marketing strategy I used 2 jobs ago? It finds the answer in seconds.
How mine works
Mine lives in plain text files in a @github repo I own, and I browse it in @obsdmd. Agents read those files at the start of every session, so they pick up where I left off instead of starting from zero. One assistant file tells any agent who I am, how I work, and where to pick up. I call it the Brain Boot file. The whole thing runs on my home server, and I reach it through @discord. There are many approaches to second brains like @karpathy's wiki and @garrytan's GBrain.
Here's what mine does in a normal week.
- When my garage flooded, it read the insurance offer, found my old receipts in @gmail, pulled comparable claims and drafted a counter. That counter was worth an extra $4,000.
- It wakes up before I do and builds a daily brief: what happened overnight, what matters today, and prep for every call.
- It triages my inboxes, archives the noise, prepares drafts and turns real asks into tasks.
- At the end of the week, it checks every meeting on my calendar against my email and drafts follow-ups for anyone I missed.
- It builds my websites, runs social media and SEO, creates my all-hands decks, builds my dashboards, prepares sales and fundraising campaigns, manages my CRM, creates videos, etc...
It can literally do anything on my computer. It can even do things on my computer I could never do, like code or advanced video animation.
Every one of those started the same way. I voice dictated into @WisprFlow how I want the job done (called creating a skill), once. Now it runs that job for me, with everything it knows about me behind it.
My fiancée told me that once she had a second brain, she felt like she was in God mode. That's the best description I've heard.
Why we moved her from Claude to Grok Bot
So back to the point of the story.
While she was running hot in god mode, her second brain had one weak spot. It lived on a server she had to babysit.
She ran the same setup I do. Claude Code on top. An Obsidian vault underneath. All of it on a Mac Mini at home, so it was always on.
Great in theory. In practice, the Mac Mini needed patching, and she had zero interest in patching it. A few times it went offline right as she left for a trip. She'd land, open her phone, and her second brain was gone.
One trip, she needed it to follow up with a prospect and send an important proposal. The Mini was offline right when it mattered.
So we pulled the plug. She went back to running Claude Code on her laptop. It worked - while the computer was open. It also felt like going backward.
So a few weekends ago, I decided to set her up on Grok Bot and ran it side by side with Claude Code.
Grok Bot is xAI's agent, separate from the @grok chat app. It gets its own computer in the cloud with files, a terminal and a browser. That cloud computer is the server she never has to patch.
The move was easy because her setup was built for exactly that kind of portability:
- Her notes, her logs, and one assistant file all live in a GitHub repo
- That assistant file tells any agent who she is, how she works, and where to pick up
- The assistant naturally logs and organizes all of her work and conversations back into her brain.
We gave Grok Bot two rules. Reboot from the assistant file at the start of every session. Write everything back to the repo at the end.
The only real cost was reconnecting her tools, which wasn't too bad. She has about 10 tools to connect compared to my 80...
Result: she now uses Grok Bot almost exclusively, on both her phone and her laptop. Her second brain still grows in files she owns. If something better shows up next year, she points it at the same repo and keeps going.
Her new setup with Grok is pretty OP...
She can talk to it. My setup is reached through Discord, so everything is typed. I can drop voice notes, but I can't have a full back-and-forth conversation. She just talks.
Grok Bot hands work off to other bots running in parallel while she keeps going.
Then there's the send card. When Grok Bot writes an email, a @SlackHQ message or a text, it shows the exact message first. You read it. You click Send. It confirms it went out, and the card folds away. I built something like it in Discord for my own setup. It works. It isn't this slick.
That's really why we moved her. She needs her second brain on the road, on her phone, without ever thinking about a server. Grok Bot gives her that.
Where I am now
So am I switching? Not yet. I still want full control: which model does which job, what's under the hood, my local models, and what leaves my machine.
But I've started experimenting. Grok Bot now runs as an orchestrator in the cloud, and it can drive my home server, which still does most of my heavy work.
I'm dual-wielding. Grok Bot in the cloud for the everyday. My home server for the heavy lifting, like automations and video processing. So far it's working well.
The part I like most is the redundancy. If my home server goes down, or my cloud subscription hits capacity, I can still reach my second brain, manage tasks, and handle email. The heavy jobs wait. The everyday work keeps going.
If you want to start your own, here's the Brain Boot file we both started from: github.com/brainboot
Follow for more about AI second brains and company brains.
