On a laptop in San Francisco, a piece of software made by two teenagers is sorting through small molecules at a rate of roughly 700,000 per second, looking for the next thing that might keep aphids off a soybean field [HuntScreens]. The company behind it, Bindwell, has decided that the slowest, most expensive part of agriculture is not the tractor or the fertilizer but the chemistry: the decade-long, nine-figure slog of finding a new active ingredient that actually works and clears regulators. Their bet is that an AI model trained on enough protein-ligand data can shorten that slog dramatically.
Founded in 2024 by Tyler Rose and Navvye Anand, Bindwell is adapting techniques from AI-led drug discovery and pointing them at agrochemicals instead of pharmaceuticals [TechCrunch, Nov 2025]. The product is a platform that designs novel pesticide molecules from first principles, using foundation models trained on large biochemical datasets rather than the human intuition that has guided most agrochem R&D for the last half century [Bindwell blog]. The company has open-sourced part of the stack, including PLAPT, a protein-ligand binding affinity model, exposed through Python, a CLI, and a web interface [HuntScreens]. The pitch to industrial buyers is straightforward: feed in a target pest protein, get back a ranked list of candidate molecules, then send the short list to a wet lab.
The market this is aimed at is unusually concentrated and unusually slow. A handful of incumbents (Bayer Crop Science, Syngenta, Corteva, BASF, FMC) account for most of the new active ingredients introduced each year, and bringing one to market is widely reported to take more than a decade and cost north of $300 million. If a model really can pre-screen hundreds of thousands of candidates per second with what the company describes as up to 30% better accuracy than existing approaches [HuntScreens], the savings show up at the most expensive end of the funnel: fewer wet-lab assays, fewer failed field trials, fewer dead ends after year five.
A back of envelope calculation: if conventional discovery costs roughly $300M and 10 years per active ingredient, and an in-silico front end trims even 20% off the pre-clinical screening spend (call that the first $50M of the budget), each program saved is on the order of $10M in cash and a year or two of calendar time. License a platform like this to four of the five majors at single-digit-millions per seat plus milestone payments on candidates that progress, and you arrive at a software business with agrochem-style economics on the upside. None of that is guaranteed, but the unit economics are at least legible, which is more than can be said for a lot of climate-adjacent software.
The round reflects that. Bindwell raised roughly $6 million in seed funding in 2025, with Paul Graham, General Catalyst, A Capital, and SV Angel listed as investors, after going through Y Combinator's Winter 2025 batch [TechCrunch, Nov 2025] [Rowan]. Graham's personal participation is unusual; he rarely angels into YC companies after the program. The thesis investors appear to be buying is that pesticide discovery is structurally similar to drug discovery (proteins, ligands, binding affinity) but has been starved of the same machine learning attention, partly because the customer base is smaller and the prestige is lower.
Seed funding raised | 6 | $M
Molecules screened per second (millions) | 0.7 | M/sec
Accuracy lift vs prior models | 30 | %
Rose is Bindwell's CEO and previously worked at Wolfram Research; Anand left Caltech to co-found the company [Bindwell website] [Forbes]. They met at the Wolfram Summer Research Program in 2023 and were 18 and 19 respectively when they started Bindwell [TechCrunch, Nov 2025] [Ventureport]. Both were named to the Forbes 30 Under 30 2026 list [Forbes, 2026]. Rose has family ties to agriculture in China and Anand to agriculture in India, which is the sort of biographical detail that matters more than it sounds: the largest pesticide markets by volume are not in California [Ventureport].
What bears will say is that agrochemicals are not pharmaceuticals, and the comparison can be misleading. Regulatory approval for a new active ingredient runs through the EPA in the US and parallel bodies in the EU and elsewhere, with environmental fate, pollinator toxicity, and soil persistence studies that no model can shortcut. Incumbents have decades of internal data that a startup cannot match, and several of them (Bayer in particular) have their own machine learning groups working on the same problem. The bull answer, which the company's first-principles blog post leans into, is that the bottleneck is not data volume at the majors but willingness to bet on computational candidates over human-curated ones [Bindwell blog]. A small team that treats the model output as the starting point, rather than a second opinion, may simply move faster through the early funnel.
The next twelve months will be about converting the platform into a paying customer relationship with at least one of the global agrochem majors, or alternatively into a pipeline of proprietary molecules Bindwell can license or co-develop. Watch for a named partnership announcement, a published validation study against a known pest target, and almost certainly a Series A in the $20M-plus range as the seed proves out. The signal that matters most is not the next round size but whether a wet lab somewhere confirms that a Bindwell-designed molecule binds its target as predicted.
The incumbent Bindwell has to beat: Bayer Crop Science, whose internal AI discovery efforts and decades of agrochem field data are the gravity well every newcomer in this category has to escape.