Ask what artificial intelligence does on a farm and you will usually get a brochure answer. Inside a controlled-environment greenhouse the honest answer is narrower, and more useful: AI is good at the jobs that involve watching a lot of numbers, all the time, and noticing the moment one of them starts to drift.
Climate control is the clearest case. A greenhouse holds temperature, humidity, vapour pressure deficit and CO₂ inside a band the crop tolerates. A human operator adjusts to what the sensors said an hour ago. A model trained on the house's own history adjusts to where the readings are heading, and it does so for every zone at once. The gain is not a dramatic yield jump — it is fewer hours spent outside the band, which is where quality is quietly lost.
Irrigation and fertigation are the same problem in a different medium. Electrical conductivity and pH drift with the weather, the crop stage and the water source. Scheduling that reacts to drift instead of to the clock uses less water and less fertiliser for the same growth.
Vision is where expectations most often outrun reality. Cameras genuinely do spot canopy stress, uneven growth and pest damage earlier than a person walking rows twice a week. What they do not do is replace an agronomist. They shorten the time between a problem starting and someone competent looking at it.
The least discussed use is commercial rather than agronomic. A grower who can forecast next month's wholesale price, and knows their own cost per kilogram, can decide what to plant and when to cut. That is a data problem before it is a farming one, and it is the reason we publish wholesale reference prices on this site at all.
It is worth being equally clear about the limits. None of this substitutes for a correctly engineered house. Automation applied to a structure with poor climate uniformity produces well-documented failure. Models trained on one facility transfer badly to another. And every claim about a percentage improvement should be read with the question: measured against what baseline, over how many cycles?
For growers who want to see what is actually available rather than what is marketed, Zekai's agriculture directory lists 167 tools for agriculture and smart farming, each scored against a real job rather than a feature list. It is the largest section in that directory, which tells you something about how much activity there now is in this corner of the industry.
At Efarms the position is unglamorous and deliberate: automation and sensing first, because they are proven and they pay for themselves; models on top of that data as it accumulates. A greenhouse that is correctly built and correctly instrumented is worth more than a poorly built one with clever software.