Two years later, it's just another Friday
I spent today wiring up something I’ve wanted for a couple of years now, and it worked. Not “worked as a demo” — worked as in it’s running right now, on a server in my own house, doing real work without me babysitting it.
What actually got built
Set up an A300 as an Ubuntu server, put Docker and Portainer on it, and got a persistent Claude Code session running inside tmux — supervised by a systemd service, so if it ever crashes it just comes back and picks up where it left off. On top of that: BMAD as the actual methodology, and a fresh Obsidian vault as the interface — the same repo doubling as both the project’s source of truth and something I can browse and tag by hand.
The idea is simple to say and was not simple to build: bring an idea to the manager in conversation, and it triages it — build it, shelve it, or point out something already solves it. Every decision gets written down. Nothing lives only in a chat window that scrolls away.
Two years, then a Friday
This is basically what I dreamt of two years ago, except it’s not a dream anymore, it’s just another Friday. Back then this was the kind of thing you’d sketch on a whiteboard and caveat with “obviously this isn’t really possible yet.” The gap between “obviously not possible” and “set it up after dinner” closed a lot faster than I expected.
I don’t think I fully believed it would feel this normal this quickly — and there’s something genuinely happy about that: not a big dramatic launch day, just an ordinary Friday where the thing I used to dream about is now just… running.
The thing that’s actually bothering me
It’s slow. Not broken, not wrong — just slow, in a way that breaks the flow of actually working. I keep wondering if it’s the model itself — is Sonnet slow, specifically, or is this just what inference at this level of capability costs right now regardless of which model is doing it?
I’d take a real trade here if one were on offer: meaningfully faster, even if the output got a little worse. Not a lot worse — just enough worse that the round-trip stopped being the bottleneck. Whether that trade actually exists, or whether “slow” is just the tax you pay for “actually works,” I don’t know yet.