AI-Powered Plant Breeding with Avalo

Mariano Alvarez, Co-Founder and Chief Science Officer of Avalo, shares how AI and machine learning are reshaping plant breeding. From genomic prediction to advanced modeling and data-driven decision-making, we explore how advanced tools are accelerating crop development and helping breeders tackle complex traits like heat tolerance, drought response, and nitrogen use efficiency.  The conversation highlights how AI is reshaping the speed, cost, and impact of modern crop development—from the lab to the field. 

Transcript

AgTech 360 discusses breakthrough technologies that are impacting growers, businesses, and consumers hear from industry and academic experts about what's on the horizon. 

Adrian Percy: So welcome back to our Climate Ready Agriculture series. Today we're joined by Mariano Alvarez, co-founder and Chief Science officer of alo, and we're gonna explore how AI and machine learning are transforming plant breeding in response to changing climates.

We'll discuss how avalos technology is accelerating the development of crops that can withstand heat, drought, and other climate stresses, and what this means for growers and the resilience of global food systems. Welcome, Mariano.

Mariano: Thank you. Thank you so much for having me.

Adrian Percy: Yeah, no, it's delight to have you, and to get started, let's talk a little bit about you.

Tell me a little bit about your journey, what got you into Ag Tech and ultimately into avalo.

Mariano: [00:01:00] Yeah. I have actually been working on plants since graduate school. I started as a sort of molecular ecologist. And then when I finished my PhD, I came to Duke, and I was mostly working on the theory of adaptation, phenotypic plasticity and plant responses in a changing climate. So this has always been kind of in my sort of research zone of interest, but I was in academia, I really thought of myself as an academic. I was applying for, faculty positions when I was finishing up my postdoc. But my best friend and now co-founder Brendan, had left academia earlier, he was in the private sector and he has this, really deep biological background. And so, after work we would go out and have a beer and I would complain about work and he would be very understanding 'cause he knows all the stuff that is going on.

And when I was finishing up he was like, hey, I think some of this research is really interesting. I think you could have real world application. I think we should start a company. And I, of [00:02:00] course I was like, yes, that sounds awesome. That's amazing. I don't really know what that means.

Um, but, you know, we went to the university, they were very helpful in, in helping us spin out the technology and the company. And we started the company in early 2020. And I think like every startup for the first year, we were just sort of two guys and a phone. So we were just calling around and trying to really do a little market discovery, figure out how people would use this technology. Learn as much as we could. Of course, I knew because I'd been working in plant biology, how plant breeding mostly worked, but it's just, it's different hearing from people in industry and other people in academia and people in public breeding programs.

Um, and then in 2021, we were fortunate enough to get into a life sciences startup accelerator in San Francisco called Indie Bio and we were basically off to the races from there.

Adrian Percy: Fantastic. So you and Brendan co-founded Avalo? What's The mission of the company?[00:03:00] 

Mariano: The mission of the company is really to just accelerate the development of new crop varieties and development we sort of feel is everything from the breeding of those varieties to getting them into the hands of farmers, to actually helping farmers grow them effectively and get that post-farm gate product onto the market. We wanna accelerate that whole process, and we're not done until that agricultural product gets out into the world.

Adrian Percy: And so I know that part of your approach is to use ai, artificial intelligence technologies, to accelerate that breeding process, which I think most people know is a very long and arduous process in normal, normal ways of doing things.

So tell me a little bit about how you use AI and how does it help accelerate? Uh, some of these very complex traits that you're working on, like resilience to, to drought and heat?

Mariano: Yeah, I mean, we started with this idea that we really wanted to work on complex traits. You know, I think we're in this amazing era of [00:04:00] gene editing and gene editing can really deal quite effectively with a lot of these somewhat more simple traits, but when you have a very complex trait with, you know, hundreds or thousands of genes that are controlling it, then you really need to try to accelerate the traditional breeding process because that's, that's in many ways, the only way you can, you can deal with those traits.

Um, so when we spun out of Duke, what we really started with was focusing on the genomic selection process and genomic prediction, and that is a way to use, uh, genetics to predict how plants are gonna do in farmers' fields, and then load the dice a little bit and pick the right ones along your breeding program and using machine learning models and using artificial intelligence to wrap in a lot more data made those predictions.

Much more effective, which was really exciting for us. That's the technology that we had, that we had spun out with and that we had started on. But since then, actually the, the idea of using machine learning and using, um, [00:05:00] predictive analytics has expanded beyond just genomic selection in our breeding programs to, um phenotyping. So now we're using models for, um, for phenotyping every year in our fields and in our greenhouses. Um, we use different types of phenotyping, so we're using different types of scanning methods for leaves and for seeds. Um, and we're also now we're using models in our operational decision making.

So when we're. We're now simulating in silico, um, breeding programs, basically a few generations in advance. And we're planning out using models, how many greenhouses we might need, how many fields we might need, um, how many seeds we might need to produce to make sure that we're hitting our timelines to get growers seed in 2027 and 2028.

So it's really become, I have to say much more expansive than I think we even originally envisioned.

Adrian Percy: I mean, it sounds like you're using it almost at every step of the r and d process and and [00:06:00] really exploiting some of these new technologies. And I'm guessing this gives you a big advantage when it comes to like time and cost of breeding new varieties.

Mariano: Yes. So I think, you know, one advantage that we have as a small company is that all of our breeding programs are new. And so we don't have anything to, you know, there's, there's nothing internally that's being upended by using these technologies. So we can sort of start from scratch. We have all of these technologies, how do we build a program that maximally utilizes them rather than just slotting them into an existing program? Um, and it does save a lot of time and a lot of costs. So, you know, an example is, this year through our, predictions, we actually, in our, on our simulations, we realized that we actually needed too fewer greenhouse spaces than we would have otherwise and so that saved us what, something like 15,000 a month. so you know, over two breeding cycles, over a year, that is a [00:07:00] pretty significant savings for a small company, by switching a lot of our labor into bee pollination and increasing, um, the amount of sequencing that we do, we actually ended up saving almost half a million in labor costs.

Um, so when we move things onto computers and away from hand labor, we really are seeing a huge amount of savings. And I also think that our time to market is a lot better too.

Adrian Percy: I mean, I, I'm sure that many of the serious breeding companies are trying to, uh, develop these tools and exploit these tools. I mean, how do you see yourselves kind of keeping ahead of the game, as it were? Um, is it that, um. Being a small company, that kind of adaptability and the, the ability to move and to, to do things quickly. Is that your secret sauce do you feel?

Mariano: Yeah, I do think, I do think it's a little bit of, you know, nimbleness and adaptability. The main advantage that small companies have that, that startups have [00:08:00] is that there's just way fewer people when you need consensus for something, and so you can make decisions much more quickly.

You know, when you adopt something new, there's not that much to upend. There's no, you know. There's not that much sunk cost,

Adrian Percy: there's no legacy activities that you are displacing or you

Mariano: Right, exactly. You know, it's, nobody has like a 10 or 15 year project that is being sort of upended by this because we, we all just got here.

Um, so I do think that that speed to, um, that speed of development and how quickly we can put something into practice is a differentiator for a vis-a-vis like, you know, bigger ag companies. And I hope that they see that too. I hope that it ends up, you know, if they come and they see our work, they can say, you know, we never would've been able to do this internally, but here this small company is able to iterate really, really quickly and we can see the benefits of it in a way that maybe we wouldn't be able to see in our own [00:09:00] organizations in a reasonable period of time.

Adrian Percy: So, so let's talk a little bit about what you're trying to solve for like stress, and abiotic stress. Um, I mean, it's very complex as you've already mentioned. I think everyone, whatever you want to call it, people know that there is huge challenges for agriculture related to climate variability and extreme weather events and climate warming and all of these types of things.

How do you see AI ultimately helping breeders? Kind of overcome these types of challenges?

Mariano: I would say there's sort of two aspects. One, one is, is more technical and I think one is more related to. Um, the business of breeding. Um, on the technical side, you know, all of the sort of technologies that, that we employ and that I think a lot of people here at NC State are employing to genomic selection and genomic selection.

All of those machine learning techniques are, are huge benefits in breeding for climate resilience, especially the ability, I [00:10:00] think, to deal with. Very, very polygenic traits like, you know, we think thermal tolerance and, and water stress are. So being able to go through the genome quickly and efficiently and say, look, I think these genes over here and these genes over here, and these genes over here, these are all important.

And then using that information to predict which crosses you need to make that is, I think, already a very, very sort of elemental part of a breeder's toolkit, and that is like sort of, it's permanently baked in now. But I think the other thing that has been an advantage of using artificial intelligence is, um, is quantifying the impact of those traits.

So, you know, it's, we sort of take for granted that, of course we need heat tolerant crops. Of course we need drought resilient crops. And that's a benefit to farmers and it is a benefit to farmers. But unlike, you know, disease resistance traits, [00:11:00] those abiotic stress traits are often hard to quantify in dollar terms because they don't come into play every year.

They're, you know, it might only happen in a hard year or it might, um, only happen in a certain geography. So using artificial intelligence to. Wrap together. The sometimes messy information that comes from the rest of the supply chain has actually been really, really helpful. So, you know, when we go to farmers and we say, you know, how often does this impact your bottom line?

That can start to get an idea of, of what the impact of drought resilient crops are for the farmer. But then we, we go to, you know, a, a cotton spinning mill. We can also say, Hey, how did this, this change in the cotton maturity, for example, or the uniformity, how did that affect your process, and they might give us data back that is not necessarily clean or nice or neat. But artificial intelligence can help us make sense of that data very, very quickly and bring that back to the breeding program and say, look, you, from the farmer perspective, you [00:12:00] might've thought that this trait was only creating, you know, $5 worth of value per acre, but actually it's creating 30 or $40 worth of value per acre because the impact that it has on the supply chain is so great.

Adrian Percy: that's a That's a, really good explanation because I have heard lots of discussion around the value of these traits and as you said yourself, maybe some of these traits are more kind of insurance For growers in only in certain conditions, they're really coming to play. But kind of looking at this more holistically and more systemically, I think is a really interesting way of showing the value overall of, of what you're trying to do

Mariano: Yeah, it's been nice, I think to be able to, expand beyond, this sort of small group. Of course, we always keep talking to, to growers because they're the first people who are handling the seeds, but they have this impact that I think sometimes even they are unaware of at, the next step and the step after that in the supply [00:13:00] chain even all the way out to clothing brands.

So in our case, you know, clothing brands are really interested in lowering their, um, their scope three emissions, their carbon footprint. And part of that carbon footprint comes from, um, how much nitrogen farmers are putting in their fields. And so in this case, you know, we can do the work.

We avalo can do the work to help translate low carbon footprint into nitrogen, use efficiency at the farm level and explain that, hey, these two things are actually the same and we're missing some of the quantification of value of these resilience traits.

Adrian Percy: So you've mentioned a couple of areas beyond kind of stress tolerance and resiliency, like nitrogen use efficiency or, or pest resistance.

you see yourselves moving into those areas as well over time? Do you think your approach will be equally valid if you like to develop traits to solve for those challenges?

Mariano: Yes, I think [00:14:00] we see our advantage as being most impactful anytime that a trait is complex, it's hard to quantify, but it has this potential for an outsized impact 

Um, so, you know, we've started really thinking about stress resilience traits, but increasingly we are thinking about things like nitrogen use efficiency. Um. Um, you know, agronomic traits that might change the, the production system. And all of those seem like a good fit for our technology because all of those traits are best addressed through an accelerated version of the traditional breeding process.

The North Carolina Plant Sciences Initiative impacts lives through innovative applications and discoveries by leveraging cutting edge research and technology. We address global challenges related to agriculture, sustainability, and human health. The.

Adrian Percy: So let's talk about crops you're working on maybe and how those products. In the field are performing so far? You've mentioned cotton, for instance. [00:15:00] So, so where are your areas of focus right now?

Mariano: So, probably about 80% of our energy and effort in 2025 went to Cotton. Um, it's our biggest program we're doing, we're running our own breeding program. We're developing our own seed varieties. We're propagating them, we're releasing them. And so that, um, takes most of our focus. But in 2025, we also started a sugar cane program and that program is a little bit early on, so it's probably only about 20% of our effort in 2025, but I think over the next couple years it'll get much closer to 50/50. So cotton sugar cane are our two, um, biggest, biggest, uh, programs. And both of those are driven very heavily by, um um interest and excitement and enthusiasm from our strategic partners who are, um, in both of those cases farther down the supply chain.

Adrian Percy: But are you engaging with growers as well and how, how do they get involved with developing [00:16:00] these different, uh, varieties?

Mariano: Yeah, so we, um, in Cotton for example, we, are really focused in.

You know, we didn't wanna take too big of a bite of the apple, so we really wanted to focus on an area that we saw that we could have the most impact. So in cotton, we've been working with growers primarily in the panhandle, but now cover a space between Amarillo and Lubbock.

And I really, I have to say, you know, I know we're an AI company. I know we do all this like sort of data wrangling stuff, but the, the most information that we get about what we should be doing in a day to day in our breeding program is really just like sitting in a pickup truck middle of a field and talking to a grower because, you know, there's just, there's lifetimes and lifetimes of, of knowledge there.

So we learned things that we might not have known ahead of time, for example, that, um, that herbicide tolerance in the [00:17:00] panhandle is actually really, really high. Um, we might not know that, you know, although there are some spacing recommendations today, as growers have shifted to dry land primarily, um, those row spacings are actually widening.

And all of those are important pieces of information for our breeding program. So we can go back to our, um, technical team and say, Hey, you know, last year you planted these really narrow rows, I think you should widen them out because that's where growers are going. Um, so that has been, has been really interesting and.

And actually the enthusiasm from the growers has been awesome. You know, I think like, especially people who maybe who, who aren't in the industry will kind of look at agriculture and say like, oh, well the growers are, you know, this, this, uh, very conservative sector. They're probably not gonna change their practices.

They're, um, they do things one way and maybe it's just sampling bias, but the growers that we work with are incredibly [00:18:00] engaged, super, super knowledgeable, more adaptable in many ways than like any other small business that you would, would come across. And we've been, I think, pretty fortunate to have, from our 2025 program, we've had an almost 100% retention rate of growers that we're working with.

Adrian Percy: Well that's great to hear that level of engagement. I think, you know, given that you are such a tech driven company, it would be reassuring to people to hear that you've got this human element, you know, still,

Mariano: yeah. 

Adrian Percy: Still being incorporated into your, into your kind of strategy and, and how you're doing things.

But looking ahead now, so what are some of the things that you are excited about in terms of projects that you can talk about? Um, crops perhaps you're thinking of working in and maybe technology advances that you are. You are envisaging for avalo in the next, you know, five to 10 years,

Mariano: Yeah.

I've been really, really enthused about our progress in cotton. Our breeding program looks really, really good. We're doing, um, significantly [00:19:00] better than, than our commercial checks. We think we have a really interesting product for the next couple of years, so I'm really excited to see that released next year.

Um, our sugar cane program has also been, um. really, really interesting. We've had a lot of engagement both from growers and from the supply chain in Australia, which is primarily where we're working on this project. Um, so that has been, um, that has been a lot of fun. I think in terms of technology, you know, there's some kind of back of the house things that I'm running the RD so I'm of course most excited about the back of the house stuff.

Um, you know, today probably, uh, 20 to 30% of our code is is AI assisted. I expect that to jump to 50% or more in the next six months, let's say. So accelerating the development of our internal technology tools, I think is just gonna accelerate, the, um, [00:20:00] capability to wrap in these sort of customer discovery conversations through AI has also been really exciting.

Machine learning and AI are really, really deeply embedded in our, our crop development programs. They're uh, quite embedded in our production, what we call our production systems, which is our interface with growers. Um, and I'm excited for those tools to also make their way into even our commercial conversations with downstream elements of the supply chain.

I think the last thing that I've been really jazzed about is, is incorporating much more autonomous autonomous um, artificial intelligence systems into the r and d process. So there's a lot of just labor intensive tasks in r and d, that are either long running or require some sort of like minor maintenance over time and those are tasks that I think are really, really well suited to the type of AI that we have available to us now, which [00:21:00] is very long running. Um, and very um sort of independently capable of managing those processes. So actually, we're not sort of planning any huge expansions to our technical team, we actually anticipate, I'm anticipating that our technical teams productivity will, you know, 1.5 or two x over the next nine months roughly.

Adrian Percy: So I always like to finish with like a big picture question and I think, no, there's probably no bigger question right now in agriculture is, you know, what will be the impact of AI technologies, whether it's machine learning or others.

Mariano: Yeah.

Adrian Percy: You know, how do you see it beyond even Avalo? How do you see these technologies shifting how we think about agriculture, how we develop agricultural technology products? Uh, is it gonna be a profound change in your, in your view?

Mariano: I have to say, I. I am [00:22:00] increasingly thinking that it will be a relatively profound change. I do think that change is most likely to happen when tools become available in a way that practitioners can interact with them. So, you know, it's all well and good for like the machine learning people to be excited about machine learning.

That's, that's their job. But I think, now we're starting to see the first sort of inklings of, physical ai, if you will, or like embodied ai , where it's going into physical products that people are using. So, um, you might have, co-pilots and tractors, or you might have, drones that are running more autonomously or that are flagging things live.

You know, these are all robotic stuff that Avalo doesn't and deal with. But I think it's so cool and I also think there's maybe like story beneath the story of how growers will end up using it, because I don't think that the first wave of technology that comes in is going to be exactly the right fit.

[00:23:00] And I think that people who work in agriculture by nature are like tinkerers often. 

So it'll be really interesting to see how this community takes apart the tools as. You know, other people have built them and then rebuilds them for their own purposes, which I think is something that people who work in agriculture have been doing since beginning time.

Adrian Percy: Mariano with that vision for the future. Just wanna say big thank you for being on the pod.

Mariano: thank you so much for having me. I really appreciate 

AgTech 360 shares relevant news and breakthroughs with audiences across the globe. Stay connected and join the conversation by following N-C-P-S-I on social media.