I gave an agent a lakehouse and control of my garden

(jatorre.github.io)

26 points | by jatorre 3 days ago

5 comments

  • zkmon 1 minute ago
    >> I travel quite a bit. Who was going to tell me whether the plants looked healthy? Certainly not my children.

    Why not? Why do you keep your family members away from your fantastic project? I would have loved to see something like "my younger kid noticed the leaf ...", "My wife watered the plants ..." etc.

    Please remove all tech stuff from your garden. Manage it as a simple human. Let other humans of your family also take part in it.

    While your project is technically awesome, you have removed all human factor from it, including yourself. You are letting the AI agents invade the last remaining turfs of the human tribes - the garden work.

    You don't need a perfect, highly instrumented garden. That was never the need. Imprecision, lack of full metrics etc are perfectly natural. That's not achievement. That's vulgarity of pouring tons of tech into something natural as a garden, just because you can.

  • ryankrage77 28 minutes ago
    I'm reminded of that adage "Given the opportunity, players will optimise the fun out of a game". A fully automated setup is cool, but if it's working well, there's not much to do. You're not gardening any more.

    I have a bunch of plants on my window sill, and considered getting moisture sensors and automating watering, but I realised that kind of defeats the point of looking after the plants myself. On top of the satisfaction of manual work, I recently realised that it's taught me an intuition - the quirks of how much water each plant wants at different times of year (even cacti are thirsty in summer!), and how soil composition and pot size affects water retention. It's probably not generally transferable knowledge, but it was fun to realise I'd come to know my plants really well.

    • asr 3 minutes ago
      Th author spent however long on setup and wrote up a blog post. They are doing something different than you, but it’s clear they are also having fun. There is room for two approaches, and I’m not going to tell anyone else their approach “defeats the point,” because we really don’t even have to agree on the point! Glad you enjoy your approach.
  • jatorre 3 days ago
    Author here. This started as a 2 m² vegetable bed and got out of hand: 8 soil probes, a weather station, a camera, and a Rain Bird valve that I control through an old Android phone tapping its own app.

    The part I found interesting: a fixed rules engine for watering kept being wrong in ways that needed context (forecast, last week's response, what the leaves look like). So an agent now runs every morning, reads the full history (4M readings, 62 MB in Iceberg), changes watering within hard limits, and logs what it expects to happen so it can check itself the next day.

    Disclosure: the data side runs on LakehouseBox, which I'm building. The code is at github.com/jatorre/garden-lake. Happy to answer anything, including "why not Home Assistant."

  • Nashooo 6 minutes ago
    [dead]
  • krishanjarvis 3 days ago
    One practical lesson with agents that control physical things: give them a small set of named actions with hard limits, such as 'water zone 2 for at most 10 minutes', instead of raw device access. Models sometimes misread a sensor value or repeat an action because they forgot they already did it. A log of past actions that the agent reads before every decision, plus a cap on how often each action can run per day, turns most of those mistakes into harmless no-ops. A plant can survive one bad guess but not a stuck valve.