A one-man company of agents that runs my work, my content, my health, and my home. From a Mac mini in Singapore.
Every agent is a Claude Code session that can read files, run CLIs, and drive real apps.
Kicks off scheduled sessions and bridges my phone to the machine over Telegram.
Agents read one shared handbook, so they never re-learn my context.
Reachable from anywhere over Tailscale. My phone in Tokyo talks to a machine in Singapore.

Reasons across 20+ projects, surfaces blockers and next actions, files my trips and inbox. It even prepped this podcast.
The whole family's labs, supplements, regimens and reminders, in an encrypted vault.
Household tax, CPF, investments and banking, sensitive figures encrypted per-domain.
Each officer is a persona over its own knowledge base. Say “act as my COS” and it takes the role.
Assembles a personalized briefing and delivers it to Telegram before my coffee.
Drafts my weekly issue in my voice, with cover art, ready to review and send.
Verify each job's output and self-heal. The monitor is a QA harness.
The demo is easy. The 8am-every-day-without-me part is the engineering.



Same foundation, three agents. I read this over coffee instead of opening five apps.

One app, many bots. I talk to my whole system by text or voice, from anywhere, on any connection.
No new app to learn, no dashboard to open. I drop a message, an agent picks it up on the Mac mini, does the work, and replies.


Most of my open-source libraries get built exactly like this: phone on a bike stand, talking to it while I cycle.
Boris Cherny, creator of Claude Code, said ~70% of his code now ships from the mobile app. I built my own version of that.
I drop what I ate, my meds, a lab report into Telegram. The Chief Health Officer files it, tracks trends, and reminds me. No paid health app in sight.


Standing in a station in Japan, on patchy hotel wifi, I fire an agent to research a place and build the itinerary. It works async and pings me when it's done.

Portfolio, blog, and these decks — all shipped by agents.
Same agentic foundation, pointed at building an audience instead of running a household.
Jobs re-run safely and self-skip if the work already exists.
A monitor checks each job's output before it counts as done.
Anything irreversible waits for me. Reminders yes, decisions gated.
Agents drive predictable CLIs, not flaky GUI clicks.
“A good QA engineer is paranoid about reliability. That paranoia is exactly what you need to run agents in production without them doing something dumb.”
Built by one QA engineer, a Mac mini, and a lot of agents. Happy to share this deck if you want it.