Bindwell

I've been stuck on something that doesn't make sense.

Pesticide use has doubled over the last 30 years.

Yet farmers are still losing the same percentage of crops to pests as they did three decades ago.

Same losses. Double the chemicals.

That's not a problem you solve by spraying more. That's a problem you solve by building better molecules.

But here's what makes this interesting: pharma figured this out a decade ago. They went all-in on AI for drug discovery. Using machine learning to screen millions of compounds in hours instead of years. Designing molecules that target specific proteins without affecting everything else.

Agriculture didn't follow.

While biotech companies were designing cancer drugs with AI, agrochemical giants were still running the same slow processes from the 1980s. $300 million and 11 years to bring one pesticide to market. Most new approvals? Just tweaked versions of 40-year-old chemistry.

Then I came across Bindwell.

Two teenagers who started this journey at just 18 and 19 years old, who walked into Paul Graham's backyard with an idea: use the same AI that designs drugs to design pesticides.

Graham told them to flip their business model. Don't sell software to chemical companies. Design the molecules yourself. License the IP.

He wrote them a check. So did General Catalyst and A Capital.

$6 million seed round.

Here's what they're building.

Bindwell uses AI to design new pesticide molecules from scratch, ones that target pests without harming bees, soil organisms, or the environment.

Let's start with the number that matters: 40%.

That's how much of global crop production gets lost to pests and diseases every year. Despite doubling pesticide use over three decades, farmers are losing the same share of crops.

Why?

Pests are evolving faster than we can create new chemicals. They develop resistance. Farmers spray more just to maintain the same yields. The cycle accelerates.

And the agrochemical industry? Stuck tweaking old compounds instead of designing new ones.

Since 2010, only around 20 new active ingredients have been approved. Most are just modified versions of existing chemistry.

Meanwhile, bringing one new pesticide to market takes around $300 million and 11 years using traditional R&D.

Tyler Rose and Navvye Anand saw this from different angles.

Rose's aunt farms in China. She dealt with pest control challenges firsthand.

Anand has roots in Punjab, where limited pesticide options directly affect crop yields.

Both were students at the Wolfram Summer Research Program in late 2023. They were working on PLAPT - a drug discovery AI model focused on binding affinity prediction.

Then they realized something.

The biochemistry is the same.

If AI can predict how drug molecules bind to human proteins, it can predict how pesticide molecules bind to pest proteins.

They pivoted to agriculture.

In 2024, they pitched the idea to Paul Graham(Y Combinator). Their plan? Sell AI tools to agrochemical companies.

Graham told them to flip it: Don't sell software. Design the molecules yourself. License the IP.

They walked out with a new business model and his personal backing.

Here's how their system works.

Think of it like designing a lock and key. You need to know the shape of the lock (the pest's protein) before you can design a key (the pesticide molecule) that fits perfectly.

Bindwell has three AI models that work together:

It predicts pest protein structures significantly faster than industry benchmarks like AlphaFold 3. This tells you exactly what shape you're targeting.

It scans 177 million known compounds to find molecules that might fit. Traditional screening would take about 242 years to check the same number.

This is their foundational model. 150 million parameters trained on over 50 billion molecular data points. It doesn't just find existing molecules. It generates completely new molecular structures that could work as pesticides.

Here's where it gets real.

They're already testing this in their lab.

Bindwell used their AI to design molecules targeting Spodoptera Frugiperda. That's an invasive pest that causes $10 billion in crop damage every year.

They ran predictions. Generated candidates. Tested them in the lab.

Result? They found peptides and small molecules that bind to the pest's proteins with nanomolar affinity.

Translation: they work. And they were designed entirely by AI.

The business model is different from everyone else in this space.

Bindwell isn't selling their platform to Bayer or Syngenta. They're using it themselves to design new molecules, then licensing those molecules to companies that want to bring them to market.

This keeps the IP in-house. And it lets them move faster than legacy chemical companies stuck in decade-long R&D cycles.

What Bindwell is really doing is bringing modern drug discovery to an industry that hasn't innovated in decades.

Pharma figured out years ago that AI can massively accelerate molecule design.

Agriculture is just catching up.

And if they can design working pesticides in a year using AI, the question isn't whether this approach works.

It's how fast the rest of the industry will have to follow.

Bindwell's blog post explaining how their three AI models work together. This is where you see the actual data: nanomolar affinity in three weeks. [8 min, Bindwell Blog]

How two teenagers convinced Paul Graham to fund AI-designed pesticides—and why he told them to flip their entire business model. [5 min, TechCrunch]

Their original protein-ligand binding model from the Wolfram program. For anyone who wants to see the actual research that started all this. [GitHub]

Two quick things:

- Enzo