A review from Mississippi and Cornell finds AI-guided lasers, robotic cultivators and precision sprayers could cut herbicide use sharply, while raising costs above $1 million and selecting for weeds that mimic crops.

Weed control has been reinvented twice in the past century. First came synthetic herbicides, then crops engineered to survive them, allowing a field to be sprayed wholesale. A review in Engineering argues that a third change is arriving, and that it will bring a new set of problems with it.

“In the last 100 years, there have been two revolutions in weed management: synthetic herbicides and herbicide-resistant crops,” said Stephen Duke, a research professor at the University of Mississippi’s National Center for Natural Products Research. “It looks to me like AI-driven precision weed management will be the next revolution. This has the potential to drastically change how we manage our crops.”

Duke wrote the review with Lynn Sosnoskie, an associate professor in the School of Integrative Plant Science at Cornell University.

Three machines, one principle

The technologies they examined, AI-guided lasers, robotic cultivators and precision sprayers, differ in what they do but not in how they decide. All three analyse images in real time to distinguish weeds from crop plants in a row, then act on the weeds alone: a high-powered light beam to burn them, a targeted spray of herbicide, or mechanical removal.

The contrast with broadcast spraying is the whole point. A conventional application treats an entire field to kill the fraction of it that is weeds. A machine that can tell the difference treats only what needs treating.

The benefits Duke and Sosnoskie identify follow from that. Less herbicide means less crop injury, less chemical exposure for farm workers, less residue on produce and a smaller environmental load.

Sosnoskie, who works with specialty crops, emphasises the crop-safety side. “One is crop safety. I work in specialty crops, and fruits and vegetables can be very sensitive to herbicide tools, which can delay crop emergence or cause damage or stunting in crops,” she said.

Growers have long accepted that cost. “We accept this amount of injury to the plant because the need for weed suppression is so much more important,” she said. “But what we’re finding is that with precision technology, we can keep some, and sometimes even all, of the herbicide off the crop.”

“This is something that could also benefit the people who work on these farms,” she said. “Less exposure, protecting workers, potentially less residue on crops and less going into the environment – there are many values and benefits associated with the adoption of this technology.”

The organic case, and the price

The clearest beneficiaries might be organic growers, who cannot use synthetic herbicides at all.

“Organic farmers don’t have good weed management tools,” Duke said. “They rely largely on hand labor, which is getting more and more expensive and difficult to find.” Lasers and robotic tillers could substitute for that labour.

Then there is the bill. Commercial mechanical and robotic systems run around $75,000 or more. Laser-based systems can exceed $1 million.

Duke relayed one case. “I have an acquaintance in Germany who is a consultant for European farmers,” he said. “He knows an organic farmer in the Netherlands who bought one of the robotic laser weeders for more than a million dollars.” Duke said the machine “saved him so much hassle that he bought a second one, but that’s not going to be the case for everyone.”

“Some of these farmers who are working with high-value crops can spend that much, but many can’t,” he said.

That arithmetic shapes who the technology reaches. High-value vegetable and fruit operations can amortise a seven-figure machine; a grower of commodity row crops on thin margins cannot, which risks making precision weeding a tool of large operations rather than a general improvement in farming.

Resistance does not go away

The review is pointed about what precision spraying does not solve. Applying herbicide to individual weeds still applies herbicide, and still exerts selection pressure for weeds that survive it.

Herbicide-resistant weeds are already a serious problem. “That doesn’t mean every weed is resistant to every herbicide, but there are a lot of herbicide-resistant weeds out there,” Duke said. “It’s become a really big problem.”

Resistance has developed in dozens of weed species worldwide since herbicides came into wide use, and in some regions growers have run out of effective chemistry for particular weeds. Using less herbicide more precisely may slow that process without stopping it.

The mechanical alternative has its own cost: excessive tillage damages soil structure, accelerates erosion and releases carbon, which is why reduced tillage has been promoted for decades. A robot that cultivates between crop rows repeatedly is working against that.

An algorithm that did not know the local weeds

A less obvious limitation concerns where the software learned its job. Many detection models are trained on specific crops in specific regions, predominantly the western United States.

“Many of these companies have very much targeted the western United States, because there’s a huge amount of vegetable production in Arizona, California, Oregon and Washington, but that’s a very different climate compared to, say, the Southeast,” Sosnoskie said.

She described a concrete failure. “I worked with a company whose original algorithm kept missing one specific weed species: common ragweed,” she said. “Well, it turns out they don’t have the same ragweed species in California.”

“That means if a vegetable grower in Mississippi, Alabama or the region wants to adopt this technology, we need to make sure it’s trained on this environment and these weeds,” she said.

Weeds that look like crops

The review’s most striking warning is evolutionary. If a machine identifies weeds by appearance, weeds that look more like the crop will be left alone, and will set seed.

Over generations, that is selection for crop mimicry, and it has happened before. Barnyard grass growing in rice paddies evolved an upright, rice-like form under centuries of hand weeding, because weeders pulled what looked unlike rice. A camera is a faster and more consistent weeder than a human, which means the pressure it applies is stronger.

“What I’ve been arguing is just because we have this technology, that doesn’t free us from weeds,” Sosnoskie said. “With weed control, we always have to think about how the weed is going to respond or try to ‘win.’ We have to be proactive with how we think about these problems.”

That framing is the review’s contribution. The question it raises is not whether the machines work, which they evidently do, but what a field looks like after a decade of them.

---

Notes

1. “Artificial-Intelligence-Guided Weed Management with Herbicides, Lasers, and Robotic Cultivators: Opportunities, Limitations, and Unintended Consequences.” Engineering, 2026. DOI: 10.1016/j.eng.2026.08.020.
2. University of Mississippi. “Researchers study how AI could improve weed management.” EurekAlert!, 20 October 2026. https://www.eurekalert.org/news-releases/1148562.

Photo by Tom Fisk on Pexels.com