Running Claude Code as an always-on second brain, reachable from my phone
Every project I've written about here -- the World Cup pool, the closet app, the SIDS simulation, InvestorCheck, the trading engine -- was built with Claude Code running as a second brain on my machine: a collaborator that holds the whole repo in its head, runs the terminal, and remembers my preferences and how I work between sessions.
It's a second brain, not a tool. It keeps a persistent memory of me: my projects, my preferences, the way I write. I wrote this post with it too, the way I write most things now: I draft and direct, it types, and I edit hard against my own past writing and a running list of phrases I won't use. The thinking and the final cut are mine. It's an extension of how I already work, not a replacement for the judgment.
Everything gets written down. I'm obsessive about saving context to project memory files and a running session log. Every session ends by recording what we did and why, so if the machine crashes or a session dies, the next one picks up exactly where we left off instead of starting cold. I also clear the context between unrelated topics rather than letting one long thread sprawl, which keeps it sharp and keeps the token bill down.
The setup is always-on, on purpose. My Mac is a laptop, which by default sleeps when I close it -- useless for a machine I want to reach at any hour. So I keep it awake around the clock with Amphetamine and leave it on. From my phone I can open a session against that machine from anywhere and drive the real thing: the actual repos, the actual terminal, live. It's the same environment I'd have sitting in front of it, not a sandbox or a mirror.
The reason it's always-on: ideas don't arrive on schedule. You don't get to choose when one shows up. It happens in line at the store, at the gym, at 3am during a feed, which is exactly when the SIDS simulation idea arrived. I used to write those down and wait until I got to a desk, booted up, and reloaded all the context, by which point I'd usually forgotten half of what I wanted to do. Now I open a session from my phone and the machine already has the context. The idea is moving before I lose it. And if it's during a feed, it keeps working while I go back to bed before the next one.
It's good at the HOW. The WHAT is still mine. I wrote a separate post about how an LLM built me a working trading system and still couldn't find alpha -- it's strong at the engineering and no help on which idea is worth building. The same split runs through everything here. Claude scaffolds the deploy, wires the test, and untangles the bug at a speed I can't match. Deciding what to make, and whether it's any good, is the part I keep.
This site and most of what I build now run through that loop: an always-on machine, a second brain that knows my context, and a phone to start a session whenever the next idea shows up.
← Writing