Plants can speak using electricity, so this startup built an AI model to decipher their language

Electric Plant, a startup co-founded by CEO Justin Mast and neuroscientist Lee von Kraus, has built an AI model called Bombadil that it says can interpret the faint electrical signals running through plant stems and turn them into guidance for growers. The company has spent about three years gathering training data and recently closed an $11.2 million seed round to bring its sensors and model to market. Onto Ventures and Climactic led the round, with Ponderosa, Endeavor Catalyst, Ludlow, Michigan Rise, The Fund, Schox VC and Dug Song also participating.
Mast traces the company to a piece of research he came across a few years ago about a houseplant wired to a motorized stand. When the plant needed more light, electrical signals moved through its living tissue, a control board detected the current, and the stand rolled the plant toward the light. Mast described that as the start of his interest in the area, and he later teamed up with von Kraus, who has worked with DARPA, to found the company.
The pitch is that growers lack real-time insight into what a plant actually needs. Mast argues that the standard toolkit — soil probes, handheld spectroscopes and satellite imagery — is sophisticated equipment but weak at reading conditions inside the plant itself, lagging behind the plant's real-time requirements. Electric Plant used those conventional devices to help train Bombadil, which Mast says can now interpret electrical signals in about 125 species. Those species span a wide range of plant forms, including ferns, redwoods, tomatoes and dahlias. Mast described it as the most diverse model built on biological electrical signals and said nothing equivalent has been done in the animal world.
The hardware is a small sensor, roughly the size of an iPhone folded lengthwise, fixed to a plant with stainless steel screws. The screws have to penetrate the cambium, the living tissue just beneath the bark, to pick up microvolt-level changes inside the plant. Readings travel over a low-power wireless network to a base station, then to the cloud, where Bombadil interprets them; results go back to the customer, who might use them to adjust watering or fertilizer schedules. Mast says only a couple of sensors per acre are needed for accurate results, and that the units cost tens of dollars rather than the thousands typical of existing tools. He also claims the system can raise yields while cutting water use by 20% or more. The company did not name its first customers.
For a developer or product reader, the interesting part is the shape of the system rather than the biology. Bombadil is presented as a large language model applied to a signal stream that is not text: continuous microvolt readings from living tissue. That implies a pipeline — sensors, a low-power radio link, a base station, cloud inference, then a customer-facing recommendation — where the model's job is classification and interpretation rather than generation. In practice, the hard engineering problems are likely to be signal quality (electrodes in wet, growing, weather-exposed tissue), calibration per species, and handling the gap between a model's output and an actionable instruction like a specific irrigation change. The company's claim of a couple of sensors per acre suggests it is betting on broad, low-density sampling over fine-grained instrumentation, which keeps cost down but puts weight on the model's ability to generalize from sparse data.
A likely trade-off is coverage versus precision. A system that infers plant state from a handful of measurement points per acre can flag trends and stress early, but it will not map variability within a field the way dense sensing or imagery can. Similarly, a model trained partly on outputs from soil probes, spectroscopes and satellites will inherit the limitations of those labels. The comparison to language models is also worth treating carefully: plant electrical signaling is a real, studied phenomenon, but translating it into specific needs at scale is an empirical claim that has to hold up across species, seasons and field conditions.
Why it matters: teams working on agricultural technology, environmental sensing or applying AI to non-text signals now have a funded example of an end-to-end stack built around cheap hardware and cloud inference, which is the pattern many physical-world AI products are converging on. If the cost and density claims hold, the practical effect is that per-acre sensing becomes affordable enough to deploy widely rather than as a pilot — an inference, not a stated result. The open question for adopters is validation: whether the model's recommendations beat existing tools in the field, and whether species-specific accuracy survives outside controlled conditions.
Based on reporting from the original publisher. Visit the source for full context and later updates.
Publisher excerpt
Electric Plant has built an AI-powered translator to help humans listen to what plants are saying with the subtle electricity coursing through their stems.