Weeds remain one of the most persistent problems in agriculture. But the biggest issue facing modern farmers isn’t getting rid of weeds; mechanical tools and herbicides can do that. Instead, the difficulty lies in identifying and killing weeds without harming crops.
Paul Mikesell is the founder of Carbon Robotics, a company that makes AI-powered robots for the agriculture sector, and the former director of infrastructure engineering at Uber. He’s spent the past six years developing AI systems that try to solve the big weed problem. His company’s solution is the LaserWeeder G2, a machine that automatically detects weeds and zaps them with a laser array.
Mikesell told Business Insider that a neural network can be important “to not just find where the weeds are, but to find the perfect place to kill the weed.” A neural network is a computational model inspired by how biological brains learn to process information, and is key to how modern AI systems function.
Across the agricultural industry, AI tools are beginning to make a difference for farmers. That’s good news for an industry struggling against foes like the climate crisis and shifts in trade. From complex robots to chatbots, farmers are testing out a range of tools to hone their processes and achieve goals once out of reach.
Machine learning takes the field
Mikesell’s experience building autonomous vehicle infrastructure at Uber helped shape Carbon Robotics’ approach to agricultural AI, applying that same technology to the farming tools he’s developing now.
The computer vision systems used in autonomous vehicles, including cars, tractors, and other agricultural equipment, often rely on neural networks known as convolutional neural networks.
CNNs are a form of neural network that can be trained to detect patterns in images. Carbon Robotics uploads images of weeds to its own database, where human labelers manually identify weeds and crops. These image-label pairs are then used to train a weed-finding CNN that can detect weeds using the LaserWeeder’s onboard cameras and computer hardware in the machine itself, meaning no internet connection is required.
John Deere, the world’s largest agricultural equipment company, also uses CNNs for multiple applications, including its autonomous tractors and See & Spray weed-detection systems. At CES 2025, the company showed its new second-generation “autonomy kit,” which can partially or fully automate common tasks, including tillage and weed removal.
Sarah Schinckel, the company’s director of emerging technologies, said AI has already improved its agricultural equipment. In 2024, she said, John Deere’s See & Spray system was used to spray over 1 million acres of farmland. Because the machine only sprays plants identified as weeds, the system was able to weed this acreage using 8 million gallons less herbicide than would typically be needed.
“If you think about that savings, as well as just overall productivity and sustainability improvements for them, that’s just a win for them all around,” Schinckel said.
The technology also gives farmers more staffing flexibility. Semi-autonomous harvesting equipment, for example, gives the human operator AI assistance that can adjust the equipment more quickly than a typical operator can react. “You can put somebody who maybe isn’t an expert combine operator in a cab, and help them still achieve high performance,” said Schinckel.
Farmers fire up ChatGPT
While big agricultural companies are building tools with complex CNNs and other types of machine learning, some farmers are making use of more accessible AI tools. Phillip Guthrie, a partner at the agriculture consulting firm Nine Creeks Consulting, often gives presentations on new technology in agriculture, including generative AI. He’s already seeing farmers pick up ChatGPT for planning and advice.
Guthrie recalled a conversation with a farmer who was having trouble with a data analytics platform he used to monitor and track weather at his farm. The analytics had never worked correctly for their operation, “so he just took the raw weather data, threw it into ChatGPT, and started doing analytics.” The AI was able to handle the analytics tasks that prior software had failed to address.
Guthrie expects more farmers to start using generative AI tools in similarly specific and creative ways, perhaps bypassing the companies that make specialized agri-tech software tools.
Two visions for generative AI in agriculture
AI techniques like CNNs, which are available today in autonomous agriculture equipment, represent a major leap in technology. Systems like the LaserWeeder G2 and John Deere See & Spray were impossible to imagine a decade ago.
However, it’s unclear how these task-specific examples of agricultural AI will fit with newer generative AI tools.
Mikesell speculated that one solution could lie in integration. Carbon Robotics, like John Deere, doesn’t use generative AI for its equipment and has no announced plans to do so. Still, he said that generative AI could become a “planning and human interface” used to operate equipment like the company’s automated laser weeders.
“I can say to the generative AI system that I want to clear this 2,000 acres,” Mikesell said. “Then, it might come with a solution and say, why don’t you deploy these laser weeders in this pattern?”
Guthrie, meanwhile, thinks that generative AI could drive a “democratization” of the industry that larger companies may well miss out on. While the industry will always need heavy equipment, he said, farmers often express frustrations with the expensive, yet extremely specific, software available to the industry. “The last thing they need is another tool that does one thing. What they want is a tool that does everything,” he said.
Guthrie said with ever-improving generative AI, “You’ll have farmers who could build their own tools, conduct their own analytics, do their own automations, and focus on what they want for themselves.”
“That, to me, changes the shape of agriculture.”
