Broadcast herbicide is a blunt instrument: treat the whole field because you cannot afford to miss the weeds. Computer vision changed the economics. Cameras on the boom classify plants in real time; nozzles fire only where weeds are.
John Deere’s See & Spray platform is the scale proof. Company and industry reporting put 2025 commercial use in the range of about five million acres, with large average cuts in non-residual herbicide versus broadcast baselines. Land-grant and extension work has separately measured big savings when application is limited to weed-present zones. That is applied machine perception on commodity row crops — corn, soy, cotton — not a greenhouse demo.
Competitors and adjacent products are industrializing the same idea: weed maps that upload to nozzle-controlled sprayers, OEM integrations with PWM control, green-on-green recognition inside an established canopy. The science stack is classic applied AI — edge models, labeled plant imagery, real-time actuation — married to farm iron that already exists.
Why it matters now: chemical cost, resistance pressure, and regulation all push away from blanket spray. Vision-guided application is one of the few AI stories where the metric is gallons not sprayed and acres already treated. For appliedscience.com, it is computer vision you can smell in the field.