I went down exactly that path with an outdoor bed: 8 soil probes, a weather station, a camera, and a drip valve. The fun part turned out to be not the automation but letting an agent decide the watering every morning and write down what it expects, then check the probes the next day to see if it was right. Fixed schedules kept being wrong in ways that needed context (forecast, last week's response, leaf wetness).

It's still humbling: it missed caterpillars on the broccoli until I made it look at full-resolution photos. Wrote it up here: https://jatorre.github.io/garden/