Teen founders raise $6 million to reinvent pesticides with AI — and convince Paul Graham to join

Two teenage founders walked into Y Combinator founder Paul Graham’s backyard with an idea that no one in agriculture seemed to want — an artificial intelligence model to help design better pesticides. By the time they left, they had a new business model, a new company, and, ultimately, Graham’s support.

Now, that reimagined company — Bindwell — has raised $6 million in a seed round, co-led by General Catalyst and A Capital, with a personal check from Graham himself. Instead of selling AI tools to legacy agrichemical giants, the startup is using its own models to design new pesticide molecules in-house and license the intellectual property directly — a shift in strategy aimed at modernizing an antiquated industry still dominated by legacy chemistry.

The use of pesticides in agriculture has doubled over the past three decades, yet up to 40% of global crop production is still lost to pests and diseases each year, according to the Food and Agriculture Organization of the United Nations. As pests evolve and resistance develops, farmers are forced to use increasing amounts of chemicals just to maintain the same crops – a cycle that damages ecosystems and accelerates resistance even further. Regulatory pressures are increasing, but most agrichemical companies still rely on modifying old compounds. Bendwell is betting that AI can break this cycle by discovering completely new, more targeted molecules, molecules designed from scratch to meet modern challenges.

Founded in 2024 by Tyler Rose, 18, and Navvi Anand, 19, Bindwell is adapting AI-driven drug discovery techniques to agriculture, with the aim of accelerating how new pesticide molecules are identified and tested.

Bendwell began as a research project in late 2023, when Rose and Anand were students in the Wolfram Summer Research Program. They initially focused on an AI model for drug discovery called PLAPT, which involves prediction of binding affinity, work that was later cited in a Nature Scientific Reports paper on cancer therapeutics. In 2024, they began exploring how to apply the same approach to pesticides.

Both founders had personal exposure to the problem. Rose learned about the challenges of pest control from his aunt, a farmer in China. Anand’s family owns farmland in Delhi, where he saw firsthand how limited pesticide options impact crop yields.

“Agriculture was on our mind,” Rose said in an interview. “That led to the realization that we could use the same technology that had been successful in drug discovery. We could transfer that to pesticide discovery, because the biochemistry is the same, but pesticides are a big problem, and I feel like they’re not focused on as much by most people.”

Rose and Anand joined Y Combinator’s Winter 2025 cohort with plans to build AI models and sell access to major agrichemical companies. But they did not find traction, as most industry players were reluctant to adopt AI as a key part of pesticide detection. Halfway through the program, they were invited to Paul Graham’s house, where they spoke with him for about 45 minutes in the backyard.

After hearing about their challenges, Graham suggested a different approach: Instead of selling kits, they could use their own models to discover new pesticide molecules themselves. That conversation marked the beginning of Bendwell’s current trend.

“Founders [of Bindwell] “It’ll probably be okay,” he later posted on X. “They’re smart and have a good idea.”

Bindwell has developed its own AI suite designed to reduce hallucinations – a common problem where models produce unreliable or unsupported output. The software includes Foldwell, a structure prediction model, a fine-tuned version of DeepMind’s AlphaFold, which is used to determine target protein structures. It also includes PLAPT, an open source protein-ligand interaction model capable of scanning every known synthetic compound in less than six hours, and APPT, a protein-ligand interaction model for biocidal screening, which has been reported to outperform existing tools by 1.7× on the Affinity Benchmark v5.5. Furthermore, the suite includes an uncertainty quantification system that identifies when results are trustworthy and when more data is needed.

“Since we’re not selling AI models, we’re not competing with companies that are selling models,” Rose told TechCrunch.

Together, the Benduel models can analyze “billions” of molecules, delivering four times faster performance than DeepMind’s AlphaFold 3, the startup said.

“The way most pesticides are detected today is target-independent,” Rose said. “Entomologists and chemists propose different compounds, then test them on insects. You often need to manufacture and test thousands of chemicals, which is expensive just to verify their effectiveness. With our AI models, you can simplify the problem down to a single protein.”

AI helps identify proteins that are unique to a particular pest but absent in humans, beneficial insects or aquatic organisms such as water fleas.

“Once you find those proteins, you can design something that binds to them and prevents them from working,” Rose said.

Bindwell is currently testing the effectiveness of AI-generated molecules at its laboratory in San Carlos. It is also working with an external partner to further validate the models, though Rose declined to share details.

The startup is in early discussions with several global agrichemical companies, and its first partnership deal is expected to close soon, Rose said. “A year from now, we want to enter into licensing agreements with some of these companies,” he said. Bendwell has also initiated talks with stakeholders in India and China to conduct field tests.

The startup currently has a team of four, and also works with external contractors to synthesize the molecules.

Bindwell’s founding tour also included participation from SV Angel, along with Graham. Before joining Y Combinator’s Winter 2025 cohort, the startup raised a seed round from Character Capital.