My House Runs on an AI Agent

People often ask what it actually looks like day-to-day to have an AI agent manage infrastructure you depend on — not a chatbot you occasionally ask questions, but something closer to a project lead that reads your backlog, opens pull requests, and pings you when it needs a decision. Here’s the actual stack, the actual hardware it runs on, and the actual daily loop, not the marketing version. The setup, in one paragraph My home runs on Home Assistant, self-hosted, with the entire configuration in a self-hosted Git repo (Gitea, running on my NAS). Claude Code — an agentic coding CLI — has full read/write access to that repo and to a set of scoped tools (deploy/reload commands, a self-hosted issue tracker, some read-only APIs for other systems on my network). It does not have access to secrets files directly, cannot restart the core service without asking, and cannot touch a handful of centrally-important files without my explicit sign-off. Everything else — writing automations, fixing bugs, refactoring, documentation — happens largely on its own, inside guardrails I’ve built up over months. ...

I’ll admit it upfront: I have a mild, ongoing addiction to trying new AI tools the week they show up, and then finding some corner of my home automation where they might actually earn their keep. Most of them don’t survive contact with reality. This is the story of one that did, and what it cost to get there. The candidate: a decision model, not a chatbot A new model showed up on my usual model marketplace — not a chat model, a decision model. Instead of prompting it with text and parsing whatever prose comes back, you send it a state description plus a set of typed questions, and it hands back a typed answer with a probability attached. No prompt engineering, no hoping the JSON parses. Small context window, and — the number that actually got my attention — a few cents per million input tokens and effectively free output. ...

Some backstory: my backlog had grown to nearly a hundred open tickets, and a fair chunk were probably stale — work that had quietly gotten done without anyone closing the ticket. A classic research task: pull up ten old issues, check them against the actual code, report back what’s still real versus what’s a “ghost.” For exactly this kind of task, I dispatch parallel subagents — small, scoped agent instances that work in the background while the main session keeps going. Three batches, ten issues each, all launched with the same instruction: ...

The introduction post gave you the elevator pitch. This one is the wiring diagram — how a coding agent ends up running semi-persistently against real home infrastructure, how I talk to it from my phone on the other side of the planet, and what happens when a session runs for hours and hits a wall it wasn’t supposed to hit. It’s not a chat window. It’s a long-running process. The mental model that trips people up first: this isn’t “I open an app, type a question, get an answer, close the app.” The agent runs inside a coding CLI, and that CLI process needs to keep existing between the moments I’m actually typing to it — otherwise every session starts from zero, re-reading the whole repository to figure out what’s going on. ...

I recently made a small change to my home automation setup: a robot vacuum that used to start on a rigid 10-minute “away” timer should instead wait until a larger geofence is actually crossed — no more triggering just because someone stepped out to the mailbox. A tiny feature. It should have been a ten-minute edit. It was not, because in this house even the vacuum cleaner has opinions about the security system. ...

I’m starting this blog for a simple reason: I’ve been running my home automation system with an AI coding agent acting less like a chatbot and more like a project lead — reading a backlog, opening pull requests, asking for sign-off on the risky stuff — for months now. It’s been genuinely useful, occasionally broken in genuinely funny ways, and at this point my AI agent has more commit access to my house than most of my actual family. ...