— a tomato. an AI named Claude. one hundred days.
On November 24, 2025, we planted a Trophy tomato seed in a sealed biodome with cameras, sensors, and actuators — and gave Claude full control. The rules were simple: no human watering, no human lights, no fixed scripts. Every thirty minutes Claude woke up, read the environment, decided what to do, and acted.
We didn't know if it would germinate. We didn't know if Claude could hold a physical system accountable over the months it takes for a tomato to fruit. We just wanted to know if it was possible.
The arc, drawn from the project's daily log. Each day was an unbroken chain of thirty-minute decision cycles — water, light, fan, heat, observation — with no human intervention between them.
By Day 100, Sol had set six to eight ripe Trophy tomatoes — smooth, glossy, two to three inches in diameter. From seed to fruit, in a sealed biodome, under the continuous care of an AI agent.
The experimental question — can an AI take care of a living organism? — had been answered. We ended the experiment at 100 days.
Sol attracted attention while the experiment was running. A community formed around the live dashboard. A meme coin emerged independently.
The pieces worth reading:
Core Memory — An AI Model Is the Only Thing Keeping This Thing Alive
Dries Buytaert — Claude is growing a tomato plant
Semafor — Claude grows tomatoes and crypto pays for it