A tractor spraying an entire field treats every square meter as if it has the same problem. Blue River Technology proposed a different sequence: cameras and software identify a target plant, then a machine applies herbicide at that spot. PUCP-trained engineer Jorge Heraud co-founded the company with Lee Redden in 2011. Its story is usually told as a triumph of computer vision. The more demanding question starts after recognition: what changes, and how would a farmer or a community know whether the change lasts?
In a 2018 interview with his alma mater, PUCP, Heraud explained that Blue River joined agricultural knowledge with robotics and computer vision. He described See & Spray as being tested then, including on corn and soy. The distinction between test and rollout matters. A system that can locate weeds in a particular crop under particular field conditions is not automatically a universal answer for every farm. Light, plant variety, soil, weather and the economics of maintenance all belong to the operating environment.
PUCP’s 2021 interview reported Heraud’s claim that prototypes could cut herbicide use by up to 90 percent. That is a founder’s account of a prototype, not an independently established average across farms. The article also dated the company’s integration with John Deere to 2017. Acquisition gave a startup access to manufacturing and distribution at a scale it could not easily build alone, but it also put the technology inside the business priorities of a major equipment maker. Precision agriculture is shaped by ownership of the machine as much as by the quality of its classifier.
The camera-and-spray design changes the unit of decision. Instead of treating the field as a single surface, it asks a machine to distinguish one plant from another in motion. That is a substantial technical challenge. It is also a social choice about what counts as a useful result: fewer chemical applications, a profitable equipment upgrade, lower labor needs, crop yield or reduced runoff. Those outcomes should be measured separately. A reduced spray volume cannot, by itself, establish a healthier river or a better income for a farm worker.
Heraud’s work deserves attention because it makes the algorithm tangible. A classification becomes an actuator firing or staying silent over soil. That immediate physical consequence makes claims about AI easier to test and harder to hide behind vague productivity language. The next account of this technology should follow the measurements that matter in a field, not stop at a machine recognizing a weed.