Turning Soil into Agriculture's Next Frontier with EarthOptics

For farmers, the benefits are clear. A better shot at turning a profit while improving their soil health. But the impact extends beyond the farm. The soil may hold the next breakthrough drug or therapy. We just haven’t found it yet.

In this episode, EarthOptics CEO Lars Dyrud sits down with Chuck Templeton to dig into why understanding soil matters, how EarthOptics is weaving AI into every part of the company, and what better soil data could mean for agriculture and beyond.

Chuck Templeton: All right. I am super excited to be here today with Lars, the CEO of Earth Optics, and I think all around just good guy. All my conversations with him have been super fun, and so, Lars, good to have you here today.

Lars Dyrud: Yeah, it’s my pleasure. Thanks so much for having me, Chuck.

Chuck: Awesome. So tell us a little bit about how you got involved with Earth Optics and what you saw in it

Lars: Yeah. I was working for EagleView, the largest aerial imaging company in North America. I was running corporate development, and one of my buddies John Dombroski, gave me a call. He had started an ag tech accelerator in Raleigh-Durham, North Carolina. He got funding from a bunch of North Carolina VCs and some of the big corporates like Syngenta and Bayer. And so they ultimately founded five companies out of that accelerator, and one of them was a company called Ground Truth that ultimately we renamed Earth Optics.

And he said, ​“Hey, we’ve got this idea for a company that uses sensors and machine learning to measure soil. But we don’t know anything about machine learning or sensors, so we wanted you to come be an advisor and help us get this company started.” And I said, ​“Yeah, sure. That’s awesome.”

I do know about machine learning and sensors. My previous company did machine learning on aerial and satellite imagery, and I’m a physicist by training. Then, about a month later, he called me. He said, ​“All right. The real reason I wanted to get you involved is to see if you’d be CEO of the company.”

And so the timing was just perfect. I was actually planning on starting a different company myself, but I’d always been really interested in agriculture and food production. Both my parents grew up on farms in kinda opposite ends of Minnesota.

So it was just an awesome opportunity to jump in and build a company from the ground up in a subject matter that I’ve been interested in and fascinated with for a while.

Chuck: Nice. I did the same thing with Aaron to get him to join our organization, too. I know that tactic. Good. Tell us a little bit about what you guys do.

Lars: Yes, EarthOptics is essentially a soil measurement company. We’re vertically integrated, so we go all the way from farm and ranch. We measure soil with sensors, collect soil samples, test those samples in our own laboratory, and deliver soil data and soil insights through our own cloud-based data systems, which are now AI-enabled.

So we serve a broad range of stakeholders that need soil information, from grocery stores and consumer packaged goods companies to carbon project developers, and we work directly with thousands of farmers and ranchers as well.

Chuck: But one of the big things you can do is actually look deep into the soil, right? And so what are you looking for when you look deep into the soil?

Lars: Yeah, so when we got started, we really thought there was a lot of improvement we could make to soil measurement. So, before we got going, soil had basically been measured the same way for the last 80 years, especially agricultural soils. So farmers would take soil samples, send them to a laboratory to test for fertility, and then they would know just about the soil chemistry at that one spot where they took a sample.

And so we felt that soil information had been underinvested for a long time. There’d been billions of dollars poured into above-ground farming technologies, and essentially no money has gone into what is literally the other half of the equation: what’s happening in soil. And so a couple of problems that we’ve sought to solve is soil is highly variable spatially.

Like anyone who’s got like a yard and grass, you can look out and see like you’ve got tufts of grass that are growing faster than others. You got a dead patch over here. Those are almost all soil variability. So you just move five feet away, you can have completely different soil chemistry.

Some cases even have different soil substructure, different materials like clay and silt ratios, and things like that. But if you go just even a few centimeters away, you actually have a completely different microbiome. And so there are a lot of challenges when it comes to measuring soil in a way that you can accurately make decisions about it.

So one problem you had to solve is the fact that soil is highly spatially variable. And so we tested a whole bunch of different sensors. We brought in satellite data sets, and we built the machine learning model that allows us to take spot soil measurements like a soil sample and build a model of how that thing is varying across a farm or a ranch and things like that.

And because we’re using machine learning, the model gets smarter and smarter, understanding correlations and variability. The second problem we were trying to solve is there are a lot more things that you need to know about when it relates to soil than just the chemistry.

It’s literally just one dimension of the problem. The soil physical characteristics are critical. If soil gets too compacted, it doesn’t hold water. It doesn’t grow anything anymore. If you’ve ever been on a nature trail, the trail itself is highly compacted, and nothing grows on it.

And so soil physical characteristics matter a lot. Compacted soils cost farmers $10 billion a year in the United States alone. But then, the last and most important piece is soil biology. Soil biology dictates just about everything in terms of how the soil functions for the way plants grow, and we hadn’t been measuring any of it. Just one teaspoon of soil has more microbes in it than there are people on Earth.

And so you think about how those microbes interact with each other and the plants. And there’s a lot that you have to unlock, and so the company pioneered soil metagenomic testing, so we DNA sequence all the soil samples that we collect, and that really has opened up an immense new world of understanding how soils function and giving us an an incredibly valuable tool to allow farmers and ranchers to be able to understand the biology of the soil, but also then understand where the biology of their soil may be deficient and then do something to fix it.

And so soil health is really about all those things, the physical, the chemical, and most importantly, the biology being in balance.

Chuck: And so what does it enable the farmer to be able to do differently? Do they save money? Do they grow a different variety of crops? Are they resistant to infestations? What does it help the farmer do?

Lars: Yeah, really, all those things that you just mentioned. Our goal is to provide data and insights that allow farmers to farm more profitably. So usually that means the combination of spending less money and growing more food, a higher yield. And our customers are usually doing a little bit of both.

Every year, farmers spend somewhere in the ballpark of, five, $600 per acre on inputs. So they’re buying seed, they’re buying fertilizer, they’re buying things like crop protection. And then increasingly they’re buying soil biostimulants, so they’re actually buying biology.

And so the question is, how are they spending that five, six, $700 per acre? And our data basically gives them data to drive every single one of those decisions in terms of how much fertilizer to buy, where to apply it more precisely, so you can use less of it.

Probably the most important decision they make is which seed they’re purchasing. So understanding the biology of the soil will help them pick a seed hybrid that is well-suited towards those soil conditions.

And so if you have a certain pest and pathogen present in your soil, you’re probably gonna wanna go spend the money on one of those hybrids that are bred to be resistant to that pest or pathogen. Similarly, if you don’t have any of that pest or pathogen in your soil, don’t spend extra money on a hybrid designed to protect against it that may not be optimized for yield.

So understanding the risks and threats in your soil is a big part of what farmers are buying and not buying things they don’t need, ​‘cause their soil doesn’t have that problem. That’s probably the most important reason they buy our product.

I’d say secondarily, understanding where they have biofertility weaknesses. A great example is mycorrhiza fungi are incredibly important to all plant growth. There are fascinating, incredibly complex fungal networks present in all healthy soils, both in the forest and on regular farms and ranches.

And because of decades of maybe aggressive tillage or of someone applying a broad-spectrum antifungal product at some point in history, we found that a good 30% to 40% of farms or ranches are severely deficient in mycorrhizal fungi. And so if a farmer can know that they can maybe affect some practice changes.

Chuck: Would they have known that in the past? Was that possible before?

Lars: Impossible to know. There was really no way a farmer could have known this. So this is the first time that somebody can actually understand. But over the last 10 years, there’s been a number of biological products that people produce where they’re basically fermenting these mycorrhizal fungi, and you can buy them, and you can apply them as a seed coating, essentially reinoculate your soil with this incredibly important fungi that is critical to the transport of nutrients to roots.

And so, yeah, if you didn’t know you needed this product, you might have bought it, not knowing you needed it, and hoped it worked. But for 60% of farmers, they have no need for this because they have great mycorrhizal structures. And so really knowing which one of those you are is critical to spending money the right way.

Chuck: Yeah. And so it’s both yield and profitability that you’re helping the farmers improve both of those, right? And so what types of percentages do you see across different farms and farmers as far as the improvement in either of those?

Lars: Yeah, we think in most cases, our customers are generating upwards of $100 of new profit per acre through some combination of higher yield and reduced

Chuck: How’s that relative to the current amount of profit they make per acre at this point?

Lars: Zero in today. The farm economy has been in pretty tough shape for the last year or so. Average farm profits are literally zero to slightly negative over the last couple of years. So yeah, its critical to be able to make those optimal decisions to eke out any amount of profitability in a tough farm economy.

Chuck: That’s fantastic. That’s awesome. We talked a little bit about the complexity of this, you said more microbes in a teaspoon than people on Earth, and obviously chemistry, biology, all that. What’s made this possible? How come it’s possible to do now?

Lars: Yeah, and the now possible is how I’ve approached the last couple of decades of my career. Probably the last 20 years, I’ve moved across a couple of different problems, and I’m looking for where new technologies or new innovations allow us to make a physical measurement that just wasn’t seen as practical until that technology or innovation came along.

And for what we’re doing with soil, there are two key things that make what we do possible and only possible in the last few years. One is on solving that soil heterogeneity problem, the fact that it’s so incredibly variable that you would need to take thousands of dense soil samples to truly understand it effectively.

We knew that wasn’t practical. And so machine learning and computation have gotten cheap enough that you can just shove tons and tons of sensor and satellite data into a system and start building pretty accurate models for extrapolating and interpolating between sparse soil samples.

So that just wasn’t practical until that kind of big data computation got cheap enough. And then on the other side, DNA sequencing obviously, just started a few decades ago, and only in the last few years has the cost of running an individual, you know, sequencer gotten low enough that it’s in the realm of practical for soil.

So just a couple of decades ago, it cost us millions and millions of dollars to sequence the first whole human genome. And now we’re doing that for a few tens of dollars many times a day at Earth Optics, sequencing soil. And so nobody would’ve even considered DNA sequencing soil ​‘cause who’s gonna spend $20,000 on a single soil test?

That would be crazy. We were able to essentially predict that, hey, at some point in time in the not too distant future, it’s gonna be very cost reasonable and only cheaper and cheaper with each passing day to DNA sequence soil.

Let’s start today, ride that lowering-cost curve, and bring value in an area people never imagined was possible.

Chuck: And yeah, ​‘cause the complexity’s gotta be ridiculous. You just shouldn’t use that much computation to do it. And so talk through the heterogeneous problem first. How is that being solved now with AI and machine learning?

Lars: Yeah, so what we do is we basically design the whole measurement system that still includes soil samples, but it’s just one part of the equation. So our technicians go to the field, and they’ve got an ATV or a side-by-side equipped with sensors on the side-by-side. They’re constantly measuring about 2 meters into the soil using electromagnetic waves.

And all that data comes back into the cloud and gets combined with every single bit of satellite data and elevation information. Any type of map or information we can get ahold of also gets shoved into the system. And then we’ve just been building and improving machine learning-based models that correlate the samples we collect in the field with other data sets and sensor data sets we gather while we’re there.

And the model gets smarter and smarter at correlating the laboratory test results from an individual soil sample with what it can predict are those soil attribute values. Phosphorus is one of the things we test for. The model starts learning how to predict phosphorus variability across that field with all the samples that are gathered on the field and all the other data sets, but with training from millions of acres of these comparisons of hey, I take a sample here, but I measure these other things that are indirectly correlated with the things I can actually take a sample with.

We collect a sample about every five acres, but our satellite data sets and sensor data sets are on the resolution of meters. So there’s just a lot of data that you can start running correlations with.

Chuck: And so talk about then the genomics and DNA and chemistry and biology, the second part of the puzzle that you’ve been working on.

Lars: So our goal is to take fewer soil samples but still understand soil at high resolution. But we still take soil samples, and when we do, our goal is to get every single bit of information that’s relevant out of that sample. And so the samples go to our laboratory. They’re tested for basically all the chemistry values that anyone would possibly care about, phosphorus, nitrogen, potassium, a long list of micronutrients.

But they’re also tested in many cases for physical characteristics, and we do physical measurement in the field as well, in terms of how compact is the soil, what’s the bulk density, how porous is it? But then we also run metagenomic testing on it. So that means we’re literally testing all the DNA in the soil sample.

And then we’ve built up a pretty decent database of DNA that we recognize as important to farmers, and some of those things are crop-specific, like pests that attack soybeans in particular or pathogens that infect only corn. But other parameters are generic soil fertility capabilities.

Like a plant doesn’t directly consume phosphorus. It needs to do so in concert with phosphorus-solubilizing bacteria, for example. And so we measure both these known pests and pathogens across a long list of important crops, as well as a pretty decent list of soil biofertility parameters that are critical for plant growth, and you need sufficient numbers of these populations there.

So we’re measuring all the DNA, and then we’re correlating to these true DNA strands of what we’re looking for. And so, by doing that, we can directly measure the relative abundance of these different microbes, or at least their different properties.

We don’t tell you exactly all the microbes that are in the soil, for example, but we know what DNA is responsible for phosphorus solubilization. We’re measuring how many of the bacteria that are in our tests had those snippets of DNA. And so that allows us to build this relative abundance characterizations of all these different biological parameters, and that is now information that the farmers can use to start driving decisions.

Chuck: Yeah, and all that sort of spatial and geolocated, and you know exactly where it’s at, how deep it is in the soil, and where different parts of the field will grow better or worse than other parts of the field. And so there’s precision application then?

Lars: Yeah. So the first side is on fertility. So we produce fertility maps at a 30-meter resolution or about 100 feet.

And typically, for fertilizer applications, we’re happy that the scale gets to 100-acre blocks. And so we’ve shrunk that by a factor of 1,000, so you can now actually fertilize every spot in the field to precisely the needs of that location and, in most cases, dramatically reduce the amount of fertilizer they’re applying, which is great for the farmer’s pocketbook, but also awesome for waterways and nitrous oxide emissions that impact climate change and whatnot.

So the first part is on fertility. And then, moving to biology, they can now put together a whole crop plan with knowledge of their soil’s biology. They can pick a seed that’s well-suited to that soil, like we talked about. They can skip purchasing something they didn’t need ​‘cause we’ve demonstrated they don’t have this common pest and pathogen.

And so it’s all about making smarter decisions with what they’re buying and only spending money on the things that they actually need to improve the outcomes for the plant and avoiding spending money on things that are not gonna help them improve yield.

Chuck: Yeah, and the thing that’s enabled this is obviously, there’s the human intelligence and really understanding how all these things work together, but then being able to use AI as a new sort of enabling tool has really just changed the paradigm of what’s possible now and what you can do as a farmer.

Lars: Yeah. We’re so excited about what AI is starting to do for not just us, but our customers as well. So we’re bringing hundreds of new data points to farmers and ranchers that didn’t have access to them in the past. But it’s not always beneficial. It doesn’t always help you to be inundated with a whole bunch of new data.

Sometimes it’s too much, and what you really want is the advice or action that data is telling you. Humans aren’t that great at sifting through hundreds of new data points to help make decisions. We usually want things kinda winnowed down to something that’s a little more manageable.

And so AI is incredibly capable at bringing these data sets together with other information that the farmer has and distilling it down into very concrete recommendations around the things that the farmer needs to take action about, like which seed to buy, which crop protection, or maybe which biological product to purchase, and when to take action in the field and expect to take action on these certain things.

And so it really goes from a bunch of data points to a literal recipe for success, and it does so for every single farmer in America. So we’re just incredibly excited about how AI can help take truly complex information and turn it into useful action for everybody

Chuck: But you haven’t stopped with just the product offering, you’ve also been very aggressive relative to many of our portfolio companies about how you’ve integrated AI into your organization, the operations, marketing, sales, finance, all that stuff.

So if you could maybe spend a few minutes as to how you’ve thought about it, from a change management standpoint, and then some of the things that you’ve been doing

Lars: Yeah. Many of us at the company, myself included, were very excited about where we saw AI was likely going over the last few years. And so it’s probably now a little over two and a half years since we’ve taken pretty significant initiatives within the company to first get the employees up to speed, and then start having that morph into very clear directives and initiatives that take advantage of this technology as it gets exponentially better.

And so starting about two and a half years ago, we basically let everybody know, ​“Hey, AI is coming. We don’t know all the things it’s gonna do yet, but we do know we wanna be at the leading edge of adoption of whatever’s happening.” And so we really started making those a top priority and required that every employee spend a certain number of hours per week learning to use AI to improve their jobs.

So before we launch any initiative, we basically say open checkbook for any AI tools anybody wants to purchase and start playing with. And then we started saying, ​“Hey, and you have to spend at least two hours per week experimenting with them, and then also reporting on what you’ve learned or achieved with that minimum two hours.”

And then it was a pretty big change management task. I’d say that we found early on that the company filtered into a third, a third, a third. A third of the people I didn’t have to tell to do anything. They were already busy spending all their spare time playing with this stuff at home.

A third of people who, with some coaching, could get up to speed and really start making use of these tools, but they need some help. They need somebody to show them how to bring some examples to bear, or whatnot. Then, almost a third were very uninterested or maybe even opposed to the use of this technology in their personal life and in the workplace at all.

And so managing that kind of general breakdown of the population was something that the leadership team spent a lot of time kind of thinking through how to handle and how to manage through. And I’m happy to say that we’ve now got that third, a third, a third is almost 100%. We’re probably sitting in the high 90s in terms of people who use AI daily as part of their job.

The technology’s advanced enough that we’ve got very concrete initiatives company-wide on how we’re using it in different departments. And so I think that the journey has morphed from what was an individual pursuit in education and kind of skill building with the employees to now being integrated across every aspect of how the company operates.

Chuck: Yeah, the whole idea of two hours per week per employee certainly starts to add up from a time-and-attention standpoint. You see these reports or whatever about, 95% of AI projects failed, and all this stuff like that.

And it shocks me that people don’t understand that there’s a whole sort of getting up to speed, understanding what the tool can do. The first time you pick up a tuba, you probably weren’t very good at it, right?

It took you a little bit to practice it and play it a few times, and whatnot. And so how did you continue to get through that as a leader, and I’m sure, when it was a little bit more clumsy, there was a little bit more hallucinating? What were some of those early days, and how did you get people to really start to push the tool?

Lars: Yeah. Probably the two biggest moments in terms of adoption, I think the company more widely recognized how valuable these tools were. We undertook a couple of initiatives. One is that we shoved all the company business data, all the information the company collects on soil, every single document we had ever generated, and all the Slack messages the company had ever produced.

We shoved that all into a single AI system and basically made this like a generic company AI that people could start asking questions of. And I think when we forced everybody to give it a try, pretty soon we started shutting down some of our other systems and requiring everybody to start building tools on this AI system.

I think once people could start asking like pretty complex questions of our company AI, I think that was a light bulb moment for a lot of folks who were like, ​“Wow, I used to have to go ask nine different people to try to track down an answer to something, and I just asked this thing, and it gave me exactly the right answer.”

So I think that was super valuable. The other thing we did that I think was important is we built a Slack channel called AI Wins, where people were asked to post like, ​“Hey, not only cut and paste your problem, but cut and paste your solution. Include the prompt that you used to get there.”

And really started encouraging people to start posting these. We gave a weekly prize for the best one voted by the company. We called the winner out on our weekly all-hands. And then we also started dedicating about a third of our weekly all-hands to AI tips and tricks and information.

So those two things happened at two very separate times, but they were clear like milestone moments for getting the company on board and helping a broader share of the company understand how valuable these tools were becoming.

Chuck: Yeah. Everyone needs to go through, and you can see when someone’s done it, that sort of zero-to-one moment, right? Where they’re like, ​“Yeah, this is okay. It’s a slightly better Google search, and then to that moment when they see the power and the opportunity. Is this an efficiency tool? Is this a capability tool? Like, how do you see it helping the organization?

Lars: Yeah. I think at the bare minimum, it’s an efficiency tool. That’s what we’re really asking of most of our employees, and our kind of division leaders are like, ​“Hey, you should not be doing repeated tasks anymore. You should be figuring out how to use AI to either write software to help you do those repeated tasks or to do them directly for you, one of those two.”

So yeah, definitely table stakes for efficiency improvement. I think over the last six months we’ve been really looking for ways to do brand-new things. I’m really asking our folks to be creative and look for new problems they can solve that weren’t even remotely solvable without this technology.

So the way I was describing it is like look for things that you can 10X or 100X using these tools, than the old way. And what are some just brand-new things? Like a great example, we’ve been working on an AI tool to write our proposals. We do a lot of enterprise customers, they have RFPs, and we just sent in a big, pretty significant proposal to Pepsi, for example.

It takes a lot of work to put those together historically. But we’ve been building an AI tool to basically drop in the RFP, train it on all our corpus of knowledge, give it a list of like 1,000 approved figures and graphs that it can pull from to make proposals with and we’re dramatically reducing just the human hours necessary to put in one of these pretty significant enterprise proposals, and that system’s just gonna keep getting smarter and smarter as we do more of it.

So an example of this 10X thing is, like, all right, now that we’ve cut down the time it takes to write a proposal, we need to respond to more RFPs. And so we started building an agent to go scour the web for RFPs that we logically should be responding to that we weren’t even aware of, to feed this machine that we’ve built that can now write excellent proposals with just a few hours of human intervention.

And so I think that’s an example of like we’re looking for doing things completely differently at just a dramatically different scale as an AI-first methodology. And marketing and e‑commerce production are another area where our marketing campaigns can be AI-driven from the start, where the AI is actually building the landing pages for the marketing.

It’s testing different campaign methodologies. It’s not doing A/B testing, it’s doing A through Z testing, and it’s done by the time you wake up the next morning in terms of zeroing in on attractive campaigns. And so we’re increasingly looking for ways that are just completely new AI-first ways of doing things.

And they may be things that the company’s never done before or dreamed of, but it’s now very practical, like an adjacency that we think we can move into, thanks to the investment cost being near zero for adding some new capability.

Chuck: Yeah, it’s less about the software and the capabilities of it, and it’s almost about the ideas, right? That’s the bottleneck, and that’s where humans can thrive, right? If you have a list of things you wanna do, you can now burn through that list or your to-do list much faster, and now you’ve gotta come up with another set of ideas which, I feel like for me, I’ve got this infinite to-do list out there, and as soon as I got through page one, I just flip it over to page two.

And that’s where I think companies are gonna win and be able to take advantage of markets. I was talking to one of our other CEOs, and they can now sell products into smaller markets that they wouldn’t have been able to service otherwise, so they have multiple revenue streams across several different areas that they couldn’t have covered before. How do you see where your business is going from, being able to have these new opportunities like that?

Lars: Yeah, there are two concrete things that we enacted this year of that example. One is that we started a consulting and data licensing services division, and it’s all AI-enabled. We feel consulting is no longer the typical SaaS methodology. So you gotta focus on the repeatable problem that you can solve, and you gotta stay out of consulting services because they’re not scalable.

I don’t think consulting is non-scalable anymore, thanks to AI-empowered tools. I think you can get the same leverage you used to with SaaS for many of these problems. And so we’re finding that there’s a whole bunch of customers we had who needed a lot of help and hand-holding.

Especially on the carbon side and climate side, we’re dealing with a lot of regulatory-based information, even non-government, voluntary, but still hundreds, in some cases even thousands of pages of documentation that you have to understand and adhere to.

And a lot of our customers needed help with that. And now helping them understand, out of these thousands of pages of documents, how are you staying current with that stuff? We finally had a lot of customers that needed help, and now our people can use AI to support those customers using a fraction of their human time to give them, essentially personalized recipes, game plans, and support documents that we would have never just bothered entering that space and these customers would have gone unhelped or been spending a lot more money with traditional consultants to solve these problems.

And then similarly on the data licensing side, we have the largest genomic database in the world. Like, not soil genomic database, but we have more DNA sitting in a database than anyone else by a pretty decent long shot. We’re now turning those databases, both the DNA database and our other soil databases, into useful insights because we can have AI go assemble licensable data packages in a way that would have taken somebody weeks and months just to prepare the data in a way that was usable to the end client.

So we’ve just completely opened the aperture on data licensing and consulting services. And so we just started that division in January, and it’s gonna be responsible for I think 15% of our revenue already in its first year. And I’d say it’s growing faster than any of our other business initiatives.

Chuck: Yeah, that’s phenomenal. I wanna go back into soil and just talk about all the great things that AI is unlocking in soil. How has soil been helpful to humans through time, and then what are some of the potential unlocks from AI that surround soil?

Lars: Yeah. So I’m incredibly excited about what AI is gonna do to help us understand soil biology in particular. And step one is to measure it. And just some fun facts for folks, like obviously, soil is incredibly important. It’s responsible for nearly 100% of all the food that humans eat.

Food and agriculture industries are about 15% of the global economy,

But outside of agriculture, for example, most of the antibiotics available to humans are sourced from soil fungi excretions. Penicillin’s a very popular, well-known example. So these were just accidentally discovered, but as we sequence all this soil information, half of the DNA that EarthOptics measures, humans don’t know what these species of bacteria and fungi are.

what they do, what their purpose is what their function and mode is. And so, as we can start directing AI to that dark half, there’s who knows how many next-generation antibiotics and cancer therapies on the human health side, helping us understand how soil microbes translate into nutritious food for us, or unnutritious food?

How can we start understanding and then ultimately programming soil biology to favorably produce the things that are important to humans,

I’m expecting AI to just, in the next couple of years, really start unlocking some of those mysteries with the kind of data that we can provide it that humans are just not equipped to be able to understand on their own, but AI is perfectly suited to be able to pore through all these petabytes and petabytes of DNA that nobody has any idea what it is.

Chuck: Yeah, it just would’ve taken decades to discover some of these things and you might come up short on something, and now some of these tools are just making those discoveries, and then obviously, the commercialization of those discoveries so much more possible.

So, how do you see that working into your business model at this point? Right now, you’re selling to farmers. You’ve got this internal effort, but how do you start to monetize this other…

Lars: Yeah. So, probably in the early days, we’ve identified some and are looking for more partners that want to use some of the datasets we’ve built to train AI to understand these genomic interactions. Understanding microbial communities is incredibly difficult and complex, so obviously one of the ways AI will help us make these discoveries is by training it on datasets we produce that include additional environmental data.

So there are more and more AI companies. Basecamp Research is a great example of one that’s popped up in the last couple of years, where they’re building themselves the clearing house of environmental DNA. So they’re looking to build a trillion-gene database. So that’s a great example.

There are now users of the type of databases that we have built. Both OpenAI and Anthropic are pursuing biological AI. Google’s been working on it with their DeepMind team for many years. And so they’re starting to become a marketplace for just the data. But we think not just using the data for training, but using the data as a discovery pipeline for everything from human therapies to industrial enzymes, and probably most directly important for us, for new agricultural products.

And so our CTO has been experimenting with creating a cloud-based research team of six different scientists, each with their own specialty and function. They think they’ve made their first discovery. They wrote a draft paper about it. But they’ve identified a handful of microbial communities associated with areas with low levels of deleterious fungi. So you got good fungi like mycorrhizal fungi. We got a lot of fungi that kill plants, eat roots, and do things like that. And so they think they have found what essentially could be the next biological antifungal product.

And so how can we get away from these chemical-based molecule-based fungicides that are pretty harmful for the environment, and even though they’re killing the thing that’s important, they usually are harming the environment in pretty significant other ways. How can we move away from those and move to biological-based fungicides, for example?

So our team of AI researchers thinks they may have found a solution, for example. So that’s just the tip of the iceberg, but it’s the kind of thing that I think will increasingly become commonplace over the next 12 months,

Chuck: That’s phenomenal. And are these teams inside, or are these partners third-party partners that you’re working with?

Lars: Oh, no, these were AI agents that I was talking about.

Chuck: Oh, AI agents. Awesome. That’s– Yes.

Lars: It was creating these internal to the company.

Chuck: Yeah. It’s one of the things I’ve seen our first couple of agents show up on org structures here recently, and so I expect to see some more here down the road, but that’s interesting. Good.

Where do you see yourself in three to five years? Like where is that frontier? ​‘Cause obviously there’s some great discoveries, and then you gotta figure out how to commercialize them and get them to market, right?

That’s the opportunity here.

Lars: I think there are two thrusts that are really exciting that I see us being able to go in. One is once you fully understand soil biology, as we think AI will, you can start programming soil biology in beneficial ways. And so that microbial fungicide I talked about, like all these things that farmers have to buy to improve yield, we can actually stop buying most of those if we can manage the soil microbial communities in an optimal and balanced fashion.

So, most problems with growing things can be solved by managing the microbial communities. And so I’m excited for the world where all the soil microbial communities for working soils, whether it’s forestry, farms, ranches, soil that humans trust or rely on for producing things that we care about, we can move from a chemical management of those soils, which is where we are today, to a biological management of the soils, where AI is actually prescribing biological changes to every single square meter of soil in a way that will optimize its performance and environmental benefits, like sequestering enough carbon to reverse climate change, for example, while producing twice as much yield as we ever had and using no chemicals.

So actually engineering, programming optimal soil biology, and that’s how farmers engage with soil.

Chuck: That’s just mind-blowing. That is so cool.

Lars: And then the second area is nature and evolution is amazing. These microbes have been evolving for billions of years, and they’ve discovered just about every single molecule that you’d ever want or need.

And so just what does the discovery pipeline look like when AI is looking at not just our agricultural soils, but all the DNA present in the world, mining it for new human, industrial, and medical use cases? We’re learning from natural microbes that exist that do these awesome, beneficial human things and isolating them and multiplying them.

Chuck: Wow. That’s very cool. Lars, thank you for your time today. Really exciting. I’m excited for the path you’re on, the direction you’re going, inspired by all the work that you’ve been doing, yes, on the technical aspect and the core differentiation of the business, but also on all the operating aspects of your business and how you have taken a lead on that.

So really cool stuff, and thank you for sharing that today.

Lars: No, thanks, Chuck. It’s been tons of fun.

Tonya: Thank you for tuning into the S2G podcast. If you enjoyed this episode, please subscribe, leave us a review, or share it with a friend.