It started, as these things do, with one light. I wanted the lamp in the hall to come on at dusk so I would stop tripping over the dog. That was the whole brief. One light, one schedule. I gave it an evening.
Three weeks later I have a Home Assistant instance, a Zigbee coordinator flashed to the right firmware, an MQTT broker, and a wall of automations that nobody asked for. The hall light does come on at dusk. It also comes on if motion is detected and the sun is below the horizon and nobody has manually turned it off in the last ten minutes, which is the sort of clause you only write after the third argument with a light switch.
The trap is that Home Assistant is genuinely good, so every small win pulls you to the next one. Get one sensor reporting and you immediately want all of them on a dashboard. Get the dashboard and you want it to do something. I went from "turn a lamp on" to writing template sensors that compute the dew point, which is a sentence I would have laughed at a month ago.
A few things I would tell my past self. Run it in a container or on dedicated hardware from day one, because you will not want it sharing a box with anything you care about once the updates start landing. Put Mosquitto on its own service and give Zigbee2MQTT a proper hostname, not an IP, because the IP will change at the worst time. And version-control your configuration.yaml. I lost an evening's worth of automations to a careless edit before I learned that, which is its own small lesson about hubris.
The honest verdict: it works, it is reliable enough that the household has stopped noticing it, and that last part is the actual success criterion. Nobody compliments the heating coming on at the right time. They only mention it when it does not. By that measure I am winning, even if the original one-light brief is now a faint memory buried under a YAML file I am slightly afraid of.
I have not turned the hall lamp on by hand in a fortnight. I will count that.