Hassan Fayed (Aydi) on Building AI for Every Farmer on Earth and Why Software Has Failed Agriculture

So one out of every seven or eight human beings work on a farm. When we shipped it, we realized about 80% of what people were asking. It was nothing related to what our product does. It was mostly related to agronomic questions. Over 95% of farms in the world never get visited by an agronomist. There's literally hundreds of millions of people who need.

Reliable agronomic advice and they need it on a daily basis. Earth today, every day will'll run a scan of every single one of your plots. It will look at the weather forecast, at your field profile, and it will send you a report telling you what you might want to do and just three months or has sort of like given out almost $50 million worth of prescriptions.

It has users in almost 200 countries. We even have one user in Antarctica. I don't know, maybe he's like taking care of some. Penguins or something from the moment they download the app till the moment they've extracted the first value moment, the average [00:01:00] now is three minutes. Has your leadership style changed over time?

Oh God, I was insufferable.

Hello, and welcome back to the FWDstart Podcast. One very quick favor before we get into this week's episode. If you're enjoying the podcast so far, please do drop a like, drop a follow, drop a subscribe, drop a download, share it with a friend, share it with a family member, share it with a colleague.

Anything and everything that you can do to amplify this podcast reach would be so, so enormously appreciated. Now. This week we have an absolutely fantastic episode for you, and I'm conscious of the fact that I'm saying that every week, but ultimately the ambition is for every podcast and every episode to be better than the previous one.

This week I had the absolute pleasure to be joined by Hassan Fayed. He's the founder of Aydi, the company behind Earth, which is an AI agronomist now live in almost 200 countries. Just a couple of months after launching Hassan, I can very comfortably say is one of the most incredible founders that I've had the opportunity to interview some insider baseball for you.

It's very rare that we run out of recording time, but on this occasion, we absolutely did. That [00:02:00] I hope is reflected in the variety of topics which we tackle. And don't be worried . If you could not care less about farming, first of all, perhaps you should, and I think that Hassan will convince you of the merit of that.

But secondly, there's just an enormous amount of value to be gleaned from this conversation. Anyway, as I mentioned, Hassan is a truly, truly fantastic, fantastic founder, and just a wonderful explainer of things as well. So. As always, before we get into it.

Warren Buffet once said, the stock market is designed to transfer money from the active to the patient, and Lebanese American mathematician Nassim Nicholas Taleb reminded us that you cannot predict you can only get ready. Those two insights define today's investing environment. You don't win. By guessing the next move you win by preparing, cutting through the noise and letting discipline compound over time.

That's exactly what Sarwa delivers. Sarwa's modern platform is grounded in simplicity and lets you trade stocks, ETFs, crypto options, and high yield savings all in one app. They keep your fees low, so more of your money goes to work with minimums as low as $500. Sarwa is loved by experts, not just beginners.

Whether you're [00:03:00] a first timer, an active investor, or a day trader, you join hundreds of thousands of users who choose Sarwa to grow their wealth as a FWDstart listener. You get $200 when you open a new account, with code FWDstart. All the details are in the description, and if you're chatting to the team, please do tell them that Jamie sent you.

So what are you waiting for? Get started in minutes and take control of your money your way. Sarwa is regulated by the ADGM Financial Services regulatory authority. The information shared in this segment is for educational purposes only, and should not be considered financial advice. Investing involves risk and past performance is not indicative of future results.

I wanted to start bigger picture. Yeah, if we could, because 2025, was a rough year for Agritech. I was reading, I think somewhere in the region, maybe 25 startups who went outta business. We had some big blowups, the likes of Plenty Insect. Now this being said, , I appreciate these are all sort of hardware intensive things that were in vogue during the, you know, the, the heady hype days of 2021.

But that being [00:04:00] said, , I was speaking , to Mark Kahn , from Omnivore before this as well, and great guy, one of the, things , that he mentioned is ultimately one of the biggest obstacles to software management with regard to agriculture and for farmers has been engagement. I suppose. What's your take on maybe why Agritech has struggled broadly over the, the course of the last number of years?

, And what's your kinda just perspective , on where things are at at present. . I think there's a misconception that farmers. Don't like technology. Mm-hmm. Right. , I think, , on the contrary, , farmers are, , very good adopters of technology because their day-to-day is rooted in science. Right. And so, but, but typically they haven't been strong adopters of software technology and there's a distinction there.

But in terms of hardware and machinery, I mean, that's basically the name of the game in terms of chemical products in terms of genetics and bio bio kind of product. I mean, you would be amazed by the complexity that exists [00:05:00] in plant genetics and crop genetics and farmers would.

Literally go travel around the world to look at new plant genetics, try them in their fields, in their climates, and if they work, they expand and they propagate. Right? So I think the reason why software hasn't really penetrated farming as a, as a, as a category of technology is because for the most part, it hasn't delivered tangible value, right?

Farmers are, they live in the real world. They like to hold things, they like to see things that give, have a clear, sort of like ROI, tangible impact as opposed to threshold. And software, by definition, it's an intangible asset, right? It's, it's something that you cannot touch, you cannot see. And so it's very difficult to make the case for, for software for a, for a farmer.

And I don't think that that has been overcome. I think that that skepticism is still there and in many cases for, for a good reason, [00:06:00] right? So what we try to do differently and our thesis is that you need to deliver value that is beyond your database and your front end, right? If you're just acting as a system of record that's not really enough value for a farmer to to adopt for an ag company.

The larger ones will implement ERP systems and things like that for, for many, for many reasons, but it's not really a premise for broad scale adoption and, and propagation. And so what we fundamentally believe is that you need to be delivering very tangible value to the farmer. And software is only a a, a.

A distribution model for that value. And so that's sort of how we approach it. And we don't try to work against the environment. So for example, farmers, for the most part, most farms don't have internet connectivity. And so how do you give them access to your solutions while they're in the field?

[00:07:00] There's a lot of barriers in terms of the experience itself. So farm workers, for example, usually they'd have like big thumbs, big fingers. So if you have small buttons, it, it's not really gonna work. If the, contrast of the colors you use , in your products are not strong enough, , your app is not gonna be legible , when the sun is reflecting on the phone.

And so there's a lot of small nuances that you can only really pick up and understand when you're. In the shoes of, of the farmer and, and on the ground or in the field. Talk to me about that then., Is this your sort of secret insight that you have from, from having this background?

Yeah. I mean, I think it's a well publicized secret. , As you said, so I grew up in a, my, my father is originally an an architect, but he sort of like went into agriculture by mistake. . Fell in love with it and built a business out of it. And I grew up, me and my brothers having to sort of like, as soon as school was out for the summer, the first month he would take us to the farm and we had to work to earn money so that we can go [00:08:00] out and the second month of the summer be able to go out and spend it.

Yeah. Yeah. So so I grew up like just working the field, working the pack house, harvesting, pruning, doing like all sorts of things. And so I really kind of like just developed a very deep and intimate understanding of, of the reality of, working on a farm and, and, and living in the field from a very young age.

That I think has been a tremendous asset in, in terms of building Aydi and, and understanding what needs to be done. And talk to me about when the idea for Aydi first comes about then, because you weren't working in Agritech directly prior to its inception. Can you tell me about when the idea kind of first crystallized for you and then the, the next steps from there?

So, so actually the idea of, I mean the, the initial premise of the idea came to me at a very young age when I was working in the field. I was like, there has to be a better way to manage the circus. What type of farming was it? Actually, we grow mainly grapes. So the same number of bunches on every tree, the [00:09:00] same number of berries on everybody, right?

So everything is like very systematic, but the human element that makes all of this work happen is quite chaotic. Meaning as much as you try to organize it. It. You have people who are from all sorts of backgrounds. For the most part, they're seasonal, and so it's not like they're employees who have a fixed schedule and they come in every day and they do the same tasks and so on.

It's every day. It's new people coming in, people coming, going out. You have to teach them and then , you get to sort of like interact with all sorts of different human qualities at once. And it's, you're out in the field, so it's like you're in this also like unmonitored, uncontrolled environment where anything can happen.

And so it was quite an interesting experience, especially when you're young and you're like a, when you're, when you're at a young age, you're essentially a sponge, so you're absorbing everything that's that's around you. And, and so for [00:10:00] me, it kind of like, I think bothered me that you, the built environment was very organized, very meticulous.

But then the, the human environment or the human aspect of it was very chaotic and, and entirely disconnected from entirely disconnected. And so every day you would have this like clash between these two different worlds. And, and so as I grew up, like one summer after the other, my thing was trying to get things organized.

That's what drove me. And I realized that the most powerful way to get things organized and to get a big group of people to, to work towards a certain goal and objective is the incentive system that you put in place, right? And so I kind of like designed an incentive scheme , for workers to try to like manage the behavior and the productivity and get the most out of what, what, what they can do.

And that was sort of like the, my first, insight into systems design and systems building, if you will. And yeah, so [00:11:00] that's where that, that kind of like small experience was where the original idea of, of Aydi came about where, where it was, , what if the small system that I've implemented in my family farm, what if it could be scaled, ?

What if this could be the norm for how any farm worker works? And it was always at the back of my mind that it's something that I cared about. But it was only during COVID when the idea came back to me in a way where I was able to sort of like piece all the different pieces together.

I remember I was in Dubai. I had another role at the time. And it was COVID and so we were locked down. It was kinda like caved in my apartment. And you have tons of time on your hand. Tons. So it was a mix of reading, stock trading video games, you know, whatever you could kill time with.

And then one day, I stumbled across this article [00:12:00] that was saying that Spain, and bear in mind, Spain during COVID was one of the hardest hit countries in the world. Like it was complete lockdown. Doesn't matter who you are, you could not go into Spain. It was very much, it was nearly after Italy, wasn't it?

Italy was kind of that first kind of where, and then it was Spain and, and there was Spain. And, and you, if you're if you're living in Spain, you can't get out of your apartment. You can't go anywhere. You can't go anything. Right. In the midst of all this very, very strict lockdown that they implemented.

I read an article where they were bringing in 30,000 Moroccan farm workers from Morocco into Spain to harvest the fields. So I looked, I remember I looked at it, I was like, holy, like beep um, go fresh. I was like, holy shit. But I mean, in a, in a country where you as a citizen are not even allowed to leave your [00:13:00] apartment, they are letting 30,000 people cross through the strait of Gibraltar, come into the country, roam free in the fields because they need to harvest the crops.

Right? And so that was like a light bulb where like, these are very important people, like critical people to. The success of society in general. It was the only people who were allowed to go out are doctors and farm workers. And that was, that was pretty much it. And the sad thing is most of these farm workers around the world aren't treated with the same reverence as you would treat a physician or a nurse or anything.

Actually, for the most part. People don't even know their names. Like as an employer, I don't even collect their names or their IDs or whatever in many parts of the world, right? And so I started, I went into a rabbit hole and realized that this like small nucleus of a problem that I saw from a, from a, from a, from a young age.

It's [00:14:00] actually much bigger than I, than I thought. I was curious about this, the extent to which you maybe dig into the history of an industry, a kinda a student of Patrick Collison from Stripe is big into this as far as really studying the, history of a space to adequately understand it, I feel like you've maybe found, have fallen into two camps.

You've like the Contextualist and then you have problem solvers from first principles. . , You very much were able to contextualize the issue as well. Yeah. You're digging back into it. Yeah. , And realize there's about a billion people in the world who work in I agriculture, which is the largest employer as an industry in the whole world by a mile.

So one out of every seven or eight human beings work on a farm. And for the most part, they are completely left out of things like health insurance, things like access to credit financial services, proper training everything you can imagine, basically. And, and they're, like I said, they're quite critical.

So I dug into this issue. . And realized that it's quite universal. And it was something that the more I dug into it, the more I realized, okay, that, that that's something that's [00:15:00] worth someone's time to at least give it a go. And I guess I was at a phase in my life as well, where I was still quite young.

I hope I still am. And I was stimulated by solving big problems that mean something to me more than anything else. And so, yeah, so I packed my bags, moved back to Egypt and decided to give it a shot at solving this, this challenge. Was your family surprised? Very much So. What, what was their reaction?

Everyone was surprised. Everyone was like, what the hell are you doing, man? You out of your mind. You're, I mean, I had a great, role and, and responsibility here in, Dubai had built some really cool things. And yeah, when I decided to move back and part of it was also like I had some a, a family situation that I wanted to be closer to my parents and stuff.

But the other part of that was also was like now, like it's now or never, if I'm ever gonna like, give it a swing at the bat, that's, that's when it's gotta be. So how do you get [00:16:00] started then? This seems like something whereby you very much need to be on the ground speaking to farmers explicitly about what they want.

Now you obviously have that domain expertise as far as you've worked in the space previously, but talk to me about when you're first getting kickstarted. Are you just, you know, going to farms? Are you chatting to founders? Are you gathering, you doing like the field work? To that end? I learned through action, right?

And so the first thing I did was I went on farms. I called up some people who I know on farm. And I was like would you be interested in this kind of solution? Whereby we basically, the idea was we tried to get a group of workers organized and provide the, the farm work as a service. Mm-hmm.

Right? And they would be like, yeah, I mean, absolutely. And so went out, got a bunch, a group of workers, got on a bus, went to the farm and start working. And that's sort of how we started. So even before building any kind of technology, it was just the brute groundwork that, that we [00:17:00] kicked off with.

And that's sort of like how I started learning everything that needed to happen. What were you initially seeing? Okay, I think that can be streamlined better or this can be more productive or we can make this more efficient? I think the number one thing was we were able to prove. That you can have a basic level of management for something like this.

So you can identify workers, you can know who's on your farm, you can collect their IDs, and it sounds like it's very, super basic, super benign, but you would be surprised by how complex of a problem it is. And you can make sure that you're not bringing in any underaged workers on the field. You can make sure that you're providing them with a basic level of training before they go out.

You can make sure that you're paying them their fair share. You can make sure that you're giving them some, for some form of [00:18:00] insurance. And so these things that seem very basic and that we take for granted in our, in our day-to-day lives. Like it's, it's inconceivable for me, for example, to go and work for a company without them even knowing my name or knowing my, or having my ID , ?

Or not paying me on time, or not paying me my salary that we agreed on or whatever that might be. And in many cases it's not because, or in most, most cases, it's not because they have some form of malign intention. It's because there is no tool or system that allows them to execute on something like that.

. And so that, I think was, was sort of like the first challenge that when we cracked it, it was like, wow , this can be done. And at the very beginning everyone was like, this is never gonna happen. Today it's become a norm. So today, like we work with hundreds and thousands of farms around the world and just Egypt, and they all now have very solid records [00:19:00] of all of their workers.

They can pay them on time, they can pay them on a weekly basis. They, all of these things can happen. And so yeah, that's sort of like how we kicked it off. So this was the initial thesis, which is very, it's nearly like hr, FinTech oriented workforce management. That's not what Aydi is entirely today.

There's obviously still a, an element of that. Was the focus initially very much on Egypt, or did you have global ambitions from the outset? No, we, we, the focus initially was very much on Egypt. And, and I think , two and a half, three years ago Egypt was hit with a pretty.

Economic challenge currency was devaluing very significantly. Inflation was skyrocketing and a big portion of it was triggered by the Russia, Ukraine, sort of where it was very funny 'cause Egypt exports a significant amount of citrus to Russia. And when, when, when the rockets started flying from, from both countries and you started [00:20:00] having the sort of like, embargoes on, on Russia and they weren't anymore on the swift system and the, like, the shipping lanes weren't shipping to Russia anymore and stuff.

You could see the impact immediately on the ground in Egypt with like citrus backhouse sort of like shutting down some packing lines because like they can't ship the fruit to, it's interesting, I hadn't really heard about it. Like it was very much, wheat was kind of the only thing maybe in an Irish context that we were really concerned about.

That's interesting. I hadn't realized the extent to which that would've impacted Egypt. Yeah. So , the impact was very sudden very powerful and clear, ? If you're on the ground in, in these areas, you could see the impact. , And on the other side as well, , we've realized that there is a, a limit to how much we can scale.

An operation or a service like that, ? Anything that relies heavily on human beings , has a limit in terms of how far it can go. So the, the limit was both operationally and financially because you, you had a cashflow deficit , inherently , in the model. That the more [00:21:00] it scaled, the bigger the deficit becomes.

And with interest rates increasing, it's much, much more expensive to, to finance this this deficit. And with the devaluation, you can't plug this hole anymore with foreign investments and capital from VCs 'cause the appetite to deploy dollars in Egypt, it dried off. Right? It dried off. And so we needed to rethink the whole.

Business model. . How does this sync up with the sort of AI moment? Are these occurring simultaneously? Is there some crossover here? Not yet. Not yet. AI was still not, hadn't it, still hadn't started emerging, right? But what we did right away is we had built this kind of like core piece of technology for us to manage the, the, the workers that we were bringing into the farms , as a service.

And so what we had realized is that farms were like very happy with the kind of reporting that we would give them and with how [00:22:00] efficient and streamlined that operation was. And they wanted to apply that across all of their labor force. And so what we started doing is we took the technology that we've built and we've repackaged it into a SaaS product and we've given it to firms.

And slowly but surely we, we started seeing that now that it's a SaaS. Business model. It's very asset lights, obviously, we're no longer suffering from the cashflow deficit and all, and the operational challenges that existed. And, and the other thing realizes that, hey, this can be applied anywhere in the world.

And so we started selling it in many countries, scaled to about 15 countries. We then also started seeing that, all right,, they're using us to manage their labor operations, but now they're asking to manage all of their operations. And so we expanded it into covering all field operations, workforce management, inventory, and it became kind of like what we call a field operating system.

Then and business was growing, going well. And then in came ai. And so I, we had a a, [00:23:00] what we did is you were obviously like we're, we're a technology company, so something like this comes along and you start playing with it, and you're. Not sure how exactly is it gonna impact you and how can you make use of it,

but what we did know is that if there's any value that's gonna come from this technology, we have to be the first to bring it to our customers. Right? Like we, we need to figure this out before anyone else. And so very early on we, we said, all right, we still don't know what exactly we should do with it, but we should definitely start learning it, ?

And so I've put a couple of our engineers purely on r and d with ai. I was curious about this, what the split was versus current product versus the RD side of things. So everyone was still business as usual except these two guys where their whole role was to play with IT, experiment, and come up with use cases.

. And obviously myself and sfa, who's our Chief Technology officer, who's like the Wizard of Oz [00:24:00] were were quite deep so we kept just playing with it for about a year. We didn't ship any product because you didn't jump the gun. You said, okay, let's wait on this. We just played with it. And then the first and most obvious use case was to plug it as an agent on top of our SaaS.

. So that it can help our users do tasks better, retrieve information and et cetera, et cetera. We did that, and then when we shipped it, we realized about 80% of what people were asking. It was nothing related to what we, what, what our product does. It was mostly related to agronomic questions, unique.

Okay. And, and, and so that was, even when we were like playing with the use cases, it didn't occur to us that this is what people will be interested in. And so when that happened and also it was we had brought on board a, a new investor. They're one of the largest agri conglomerates in the region TX and [00:25:00] Ibrahim Nara, who's the third generation in the, in the family business.

He was also very bullish on the agronomic piece. , What kinds of questions were they asking? So things like should I use calcium nitrate or calcium format for my for my for my field? So what kind of fertilizer should I use? What kind of past should I use? How do I calibrate my machine?

For the spray. Should I irrigate this, this or that? What kind of variety of a certain crop? Should I be planting? So anything and it wasn't configured for that. No. That interface. This was purely just people were going off script. No, people were just like, 'cause I think that's the power of, of the interface, but I think that AI took off in the way that it did.

Because of the chat experience that it, that chat GPT came out with where it's basically like a blank sheet of paper and they've given it to people and do whatever you want with it. And, and that's what, I [00:26:00] think that's one of the main factors why it was propagated so, so, so quickly. Not because when it first came out, the quality of the answers were necessarily very good, but because it was such a familiar experience that had very little limitations that allowed a lot of people to just experiment with it and, and see what, what's kind of like sticks.

So we, we, so I think it was like the writing was on the wall. That agronomy demand for agronomic knowledge is very high. And obviously started digging deeper into that and go and going back into that figure of a billion people working on farms. agriculture is a science. And that's something that a lot of people don't really understand.

They think that, oh, like watering the plant. And yeah, think of manual labor. They don't think of what's going on behind the scenes. Chemistry, you're, you're like digging holes. You're putting seeds, you just water and, and then you harvest and that food's on the table. And that's pretty much it. But in reality, it's a very, very, very advanced and [00:27:00] complex field of study and, and science.

And the stewards of the science are called agronomists, right? And so these are people who go into university. They study plant biology. They study crop science and they graduate and they become either agricultural engineers or agronomists if they go deeper into their fields of study. And agronomists are a very rare breed.

There's very, very few agronomists in the world. And the really good ones are even more scarce. And think of them as physicians or doctors for your plants. So they are the ones who go to your farm and will tell you this plant is good for you. It can work in your climate, it can work on your soil or not.

This is how much, this is like your seasonal plant. How much fertilizer you need to apply, how much different chemicals you need to apply the schedule of your operations, et cetera. And then they would routinely come to your farm and update these plans and these [00:28:00] guidelines based on the weather, based on how your soil is reacting to, to everything based on how the plant is growing and so on.

. Because plants are a living organism. It's not a production line where you just set something and, and you input equals output. No, it's, there's a lot of variables in play. And how frequent an engagement would a farm have with an agronomist, for example, like, this is interesting. I'm drawing a parallel to like telemedicine now, which I hadn't previously.

In that it's solving this problem whereby, especially regionally, there isn't enough doctors per capita or even for mental health as well, which isn't really something that I thought about, but , is this shortage of agronomists quite acute? It's. Incredibly acute. So over 95% of farms in the world never get visited by an agronomist.

Wow. Right. And so there's, there's 5%, which are mostly the commercial farms are the ones who have enough money to bring in an agronomist and, and get, get this kind of expertise and who have [00:29:00] also their own in-house agronomists and so on. And then for the other 95%, what they would get is through the fertilizer companies or the input suppliers or the seed companies, they would send their agronomists who are actually sales reps.

. And so they're pushing a certain product onto the farmer. And so the advice that they're giving the farmer is not always, it's biased. Yeah. Right. It's biased. And, and, and there's a certain economic incentive. Behind it that's not aligned with the farmer's success. And so even in, in, in the rare occasions where they do get some agronomic advice, it's not always reliable.

So too big pharma and big agronomy. Yeah. That's something I'd realized before this. Yeah. There you go. . So that's kind of like , the premise, we've realized there's a significant gap. There's literally like hundreds of millions of people who need reliable agronomic.

Advice and guidance, and [00:30:00] they do not have access to it. And they need it on a daily basis. It's not a, something that you would look up once a year. It's literally every single waking day, you're making dozens of micro decisions on what you want do, on what you're gonna do with your crops. And if one of them goes wrong, it can have severe consequences on your crop, on your livelihood, on a lot of things.

And so it's, it's also like a very important field of study that needs to be managed carefully. So when we realize this, we, we also like realized that the architecture, the infrastructure, the experience, even to a large extent, , the way the entire business has been built is not well suited.

To serve that kind of demand or that kind of product. And so we needed to sort of like re-architect the whole thing. And so we've built Earth as a standalone product. And we've launched it in November. So about, almost three months ago now. Now it's [00:31:00] in, in three months, it's now available in or it has users in almost 200 countries.

So almost in every single country in the world, we even have one user in Antarctica. Wow. That's something I'd love to know what they're doing. I would love very surprised. Like but yeah, even a, a user in Antarctica which was interesting. I don't know, maybe he's like taking care of some penguins or something, but I was gonna say, I dunno if they, I don't think they even have a VPN to go to Antarctica, so that really is someone under the ground in Antarctica.

He was curious. Yeah, yeah, yeah. Over now close to 80,000 users on the using the product or is, has given close to 250,000 prescriptions. In terms of advice and just to calibrate like a prescription like that would've required an agronomist to go and visit. So you could say on average the cost of a prescription is $200 across the world.

So in, in just these three months worth has sort of like given up almost $50 million worth of prescriptions. That's kind of like the [00:32:00] economic value that has created. And the reason why I'm, I'm sort of like mentioning these, these numbers. It's because it, I have never seen anything close to that in terms of like the speed of impact that a technology can have on, on a very diverse group of people.

And it sort of like just speaks or gives a glimpse on, on what, aI , can deliver a lot of questions to, to unpack here. Go for it. The first thing is, you mentioned there the importance of the advice or prescriptions being accurate or correct, and how crucial it is that it's not giving misleading information that ends up in a situation whereby in prior crop fails.

Talk to me. This is a very specialized field of study. How do you go about training the model then? What does that look like? , So if you think about it, there's, there's sort of like three, three steps to to getting to the output. There's the inputs, which is [00:33:00] what data are you giving the model as, as, as context for it to understand?

And that's what we call environment building, right? So what environment are you having your agent live in? What can it see? What can it read? What can it look at? And, and, and, and that's a very important piece of context. The second is the model itself, right? And so the what model are you using and how are you fine tuning it?

How are you training it? What are the set of instructions that you're giving it? And then the third is the output. How are you delivering that answer or that recommendation to the user so that it's actionable, valuable et cetera, et cetera. And how do you also track the outcome? Of that recommendation so that if you've given a good one, you know that it's a good one and you can repeat it.

If you've given a bad one, you can learn and avoid making that mistake again in the future. So what we're initially, right now at this stage, super focused on is the environment, is we're focused on [00:34:00] giving earth as much context as possible through all sorts of sources that we can gather and that the user also can provide for orth so that he understands his context.

And so form profiling and field profiling is. Very important for us at this stage, and this is where we're focused and we're expanding the set of tools and data points that Earth can have access to from a farm with the minimum amount of input needed from the user. Right? So we're trying to rely as much as possible on remote sensing technologies, satellites.

If they have already existing softwares or databases, they can easily plug them into these into, into Earth and give it that context plus. Whatever they can give to Earth on a daily basis from the field that gives them sort of like a real time view on what's happening on the ground. The second phase is on the models.

And here there's, you use, you [00:35:00] use a set of different set of models from different frontier labs depending on the use case. They work within an agentic workflow that we've crafted. And there it's very much about calibrating the, the, the rules and the instructions for these agents to make sure that they are, we minimize, I think that what we've, what we've noticed is that hallucination used to be a problem.

It still is, but to a much lesser extent. I think that current generation of models is quite good. I was gonna say, is that due to a step change in Yeah. A recent model? Yeah. Is there, do you have a preference for a particular model? Honestly, every time I say I have a preference, it changes. It changes a month later.

Yeah. And so it's, it's, it's just moving so quickly. I'm someone who's like day to day is living with these models and understanding these models, and I have a lot of trouble just keeping up with them. It's just the, the pace of, and the thing is you read the announcements, like for example, oh X just dropped a new model.

. Or open [00:36:00] AI just dropped a new model. , What we don't know is that even within the models that are live, they're still doing updates to these models that they don't necessarily announce They're constantly doing enhancements and improvements that they're not announcing as a new model.

But you can see it in the behavior, especially like we are on our own evals, so we have an evaluation framework for all the models, and so we would run an eval, an eval on a model, and then a month later we run it on the same model. And the outcome is different despite there being no external announcements that something has changed despite Exactly.

But you can see that there has been things that have changed. And, and you sometimes you can feel it. It's like when you use these models enough, it's as if you, you start and it's, it's a bit ChatGPT Definitely. I definitely notice that I maybe less so on a Clara Gemini, which attribute. Definitely.

It's like a personality shift overnight sometimes. Exactly. And it's like exactly what's going on. I was talking to someone else yesterday. Exactly. And, and, and the idea that models have a personality is, I think is absolutely true. Dario Amodei, who's the founder [00:37:00] of Anthropic, just published an essay last night.

I was up all night. Oh, I haven't heard this. Reading it. So he published an essay about a year ago. Oh, this was like an internal essay, was it? No, no, no. Just separate. No, we published externally. So a year ago he published machines of Loving Grace, which was kind of like the, the bull case for, for, for ai.

And he just wrote another essay. That's called the adolescent, I forgot the title. Have it, we linked to adolescence of of AR or something. Something like that. Yes. Something has to do with adolescence. Okay. Okay. And, and kind of like sharing and he, it, this is kinda like more of the, like the, I wouldn't call it the bear case, but sort of like the how would shit hit the fan kind of version.

So more circumspect. He's kinda like, hang on. Okay, let's, yeah. And, and he was, he, he named out an example that I was stuff, I was like, this is, this, this is quite creepy. Where Claude in, its in its training basically tried to blackmail a fictional employee at [00:38:00] Anthropic because it believed that that employee wanted to shut it down.

, And why, why did that happen? It's because they, they tested in a lab environment to give it, in the pre-training, to give it some data that would give it the, that that would make it understand that it has to survive and that humans are out to get it right. So they, they gave a bunch of like science fiction movies and, and, and novels that would, where the humans are sort of like defeating the ai.

And that was like a core part of its pre-training data. And so it kind of like had this like fear of humans and this understanding that humans are out to get me and so I need to get them before they get me. And, and so, so it's, it's concerning when you think of a, you know, it's not maybe something that we've seen yet, but you know.

Sovereign, maybe bad actors influencing particular models somewhere in the world to , subtly insert a particular perspective or viewpoint. And he was certainly more muted at Davos this year. I felt, I dunno if you saw, he did like a fireside chat with them as a SaaS as well, and he was [00:39:00] definitely a small bit more, not quite as bullish.

So this, this perhaps explains it. I think generally, like we're underestimating the, the level of sort of like society defining questions that we need to answer as a, as a species. I think that's fair. Yeah. And not to play into any dystopian scenario, like just I just give you an example of what we're doing and how we're applying AI and the benefit that it's giving to a lot of people at breakneck speed.

Mm-hmm. But I think that there's always like two sides to an equation. And I think that the other side warrants a discussion. I think that, and, and, and I think it's more of a, like in the US they're obviously having these conversations to some extent. I think probably in China they're thinking about these things as well.

'Cause they're very, probably like very well organized in terms of how they think about the long-term impact of technology and how to use technology to their advantage. I think in, for most other countries in the world, [00:40:00] including Europe, I think people are just putting their heads in the sand. And, and, and I think it's, people need to wake up to what's happening and, and, and where do you as a, as a, as a country or as a society are gonna end up being in the, in the food chain after this?

Because every single big, like technology shift creates winners and losers. And this is, I think has the potential to be the biggest like. Technology shift that, that we, that humanities has ever seen. And so the outcomes of it are gonna be much more acute in terms of like the divergence of, of outcomes.

What's the North star metric for you with regard to use case? I was listening to podcasts yesterday, was like Max from Lara um, Lara and Harvey and the legal tech space. And they're talking about, , the number of queries per day that you're asking , the model or time spent, what's the north star , for.

, For us it's the value of prescriptions. Net worth is giving out [00:41:00] recommendations. So we try to quantify how much would a farmer have paid to get that piece of information from any other way. And we're trying to maximize that. So we're trying to maximize, and for the next year, that's all we're focused on maximizing the value that we're giving out to farmers.

Right. So that, going back to that initial point about feeling a tangible ROI, that's, that's, that's how we wanna make sure that we have a, a good product in the market is by making sure that we're giving out a lot of value and then later on we'll be able to, to generate value , for the company.

But I think first we need to really perfect that value delivery piece. So that's what we're focused on. How have you overcome maybe initial skepticism? Because I think one of the differences, you know, a C-suite executive working for a law firm or a big enterprise. Need to be seen to adopting some sort of AI technology.

Nearly not the case with farmers. ? . Was there initial skepticism , from farmers that you were working with? A hundred percent.

And there's always, I think [00:42:00] generally, and I wouldn't just frame it as from from farmers, any, any, most, most people, I mean, have a, have skepticism from new technology. What we've done, I think very differently with Earth is our go to market strategy is completely different from what we had done previously.

So before we would go and speak to companies, try to find a decision maker, try to build a case, you know, try to convince them to buy our software. Can I ask what size of company are we talking about maybe initially, in terms of who your ICP was initially we were, we were looking at export oriented large scale farms. So typically over a thousand acres of land at minimum. We would have smaller ones, but the, the bulk of it were companies that have tens of thousands of, of acres, thousands of workers to manage. So like quite a big . Big operation. , We flipped the playbook completely , with Earth.

We're going straight to farmers of all sizes. We're giving away a lot of product for free. So we're having them just be able to download an app or log into earth through [00:43:00] web. And through a very, very simple onboarding experience. Just immediately start getting value and then continue using the product for free.

And like at. L at a very like, deep level of engagement, do they then start hitting paywalls? But for now, like I said, we're mostly focused on, on, on giving away as much value as possible. How do you reach a farmer? Well, there's a lot of ways to reach a farmer. You're gonna Facebook ads you, right? Like what's the, I mean, I mean, definitely there's a big element in terms of social.

. There's a big element as well in terms of on the ground activities through farming associations, through trade shows these kinds of things. So yeah, there is a combination of, of channels that we use. Have you found any one channel Surprisingly useful? Yeah. Thus far? Yeah. Go on. Go on.

Yeah. Do you share? You're like, no, I think, I think, I think I think social is really is, is is quite an interesting avenue. Yeah. [00:44:00] I'm surprised by that. Yeah. I was, you know, I was being facetious when, when I mentioned, you know, something like a Facebook ads or, and Instagram ads, but they're actually, you, you found that surprisingly effective.

Wow. Okay. . And you've been geographically indiscriminate as far as earth? Or is there particular countries or locations which are very much so, so, no, we're, we're, we're very much focused right now on certain countries in South America. And that's where we're investing some of our marketing activities and, and, and, and so on.

All the others are coming in organically. Okay. We we're not investing dollars in Antarctica. Not yet. So we're, we're, we're, we're focused on, on a few markets in, in South America for, for the time being. And then from there we plan to go into the more developed markets, so to speak, or we're just waiting to get the product to a certain level of maturity before we go into these markets quite aggressively.

What is it about South America that makes sense? I know we're kind of coming full circle there from the start or, or detour to Peru where we're coming back to. What is it about South America? Latin America, it's the [00:45:00] largest agricultural market in the world in terms of production in aggregate. If you look at all the countries together it's quite uniform.

Most countries speak Spanish except for Brazil speaks Portuguese. And so Spanish and Spanish and Portuguese can go a very long way there. The. Industries there have, are, have become very sophisticated. And so users are quite sophisticated. And there's a very strong demand for agronomy and crop science.

And for the most part they're export oriented. And so they care a lot as well about compliance and things like that, that Earth can help with in terms of like, understanding based on the US' FDA standards. What products can they use in the field? What products should they not use? And so on and so forth.

So that's a big use case as well for us in, in, in South America. And then the unit economics are much more forgiving than Europe or the us And so to some extent you can experiment much more [00:46:00] freely with different go to market strategies in South America versus the US or, or Europe. And to a large part, once you've cracked something, it is quite transferrable to the US and to Europe, because keep in mind most people who work on farms in the US or Mexican so if you crack Mexico, you crack the Mexican behavior, culture, mindset, et cetera.

It's a, you've, you've sort of like, you're 50% through the us in terms of what, what could potentially work, at least in theory, how localized is the offering then? Not super localized. Really nothing changes from one country to another. And that's, again, I think what AI is changing a lot is, is you, you don't need to localize, you just need to make sure that you're giving it as much local context as possible.

And, and here it's not about the country, it's about that specific [00:47:00] user or that specific field in our case. And that's how you win. But then in terms of language, so even, I mean our, our system languages are mainly English, Spanish, and, and, and Arabic. But we have people speaking to Earth in Russian in Mandarin.

And it will respond in that language. So yeah, I think what we might at some point need to, localize on, on the payment solutions to make it easier for people to pay in their own local currencies with the, with the stuff. 'cause Stripe only goes so far, especially with farmers. But besides that I don't think there's gonna be a hell of a lot of customization to do.

How difficult is it because the team is so geographically distributed then at a relatively early stage? Like you're in Dubai, you've Cairo, you've Madrid, you've Peru. Is it difficult, even just from a time zone perspective, , there's a lot of fragmentation there. How do you find managing all of that?

If, if anything? I mean, I think it's so I, for one, I, I cannot [00:48:00] work remote, for example. This is what I wanted ask. I, I hate working from home. I cannot, I need to wake up, take a shower, have a coffee, have breakfast, and get out and get out. You know, like, and I need to be around people. I need to communicate, feel the energy.

I dunno, I feel like I'm an old school now. No, totally. It's not that I even wanna have a conversation with you, but I'd like to sit next to you and if you're talking to someone else, that's sufficient, that I'm absorbing a hundred percent. So I think that definitely that element is lacking and, and my advice to anyone who's kinda like early on in their career, especially now with what's happening with ai.

If you want any shot at Success in life, do not work remote. Really feel that strongly about it. I, I'm, I'm, if, if your job can be done remotely, there is a very high chance it'll be done by an AI in the not so far away future that will be available 24 7. That will be much more competent and [00:49:00] that will be a hell of a lot cheaper.

What will. Make humans be valuable is the human side. And that only comes by interacting with people and, and, and being physically present in an environment with people. So I think very, I feel very strongly about that. Having said that, our setup, I think the only reason why we're, why we're able to build a product today that is on par with the best funded, largest global technology companies in terms of quality of a product is because we have such a diversified talent pool.

And so Edward, for example, who is our chief product officer, he sits in London. He was a director of product at Spotify. He was head of product at Amazon. Bosco was our chief growth officer, sits in Madrid. He was head of growth at gorillas and at WeWork and PayPal and some of these.

So by, by having this diversified pool, we were able to get. Very, very [00:50:00] high quality talent that is able to that, that allowed us to build a, very high quality product. So I think e to weigh the pros and the cons, but what we try to do is, like, for example, Madrid, everyone's working at the office in car, try to bring people at the office as much as possible and so on.

So we have these kinda like microcultures and we merge together in the virtual realm. And how did you hire these guys? How did you come into contact with them? Through people who introduced us to each other when I was looking at them. I mean, I'm, I'm, I'm, I think this has, this has been my greatest blessing in this whole journey, is just finding incredible people that I enjoy working with.

'cause building a company is such a lonely endeavor and in many cases it's incredibly difficult. And like mind chattering. And when you have people around you who are just fun to be around, that goes a whole, that goes a very long way. Like when, when, when, when, when the room turns very somber and then [00:51:00] someone suddenly just cracks a joke and everyone just like, you know, let's off And that, that it's these kinds of things.

I think that, that make a journey or an experience so much more worth living. And you're a solo founder as well. Am I correct in that you don't have a co-founder? I, I don't, I don't have a co-founder like, officially, officially. But, but I do believe that like the whole team are, are like, I, I look at myself as sort of like having a role within a broader team.

But there's everyone else in the company is so much better than I am and what they do and. My role, I have a very specific role in the company and I try to keep at it. And, and then I, whenever there's a bit, there's like a problem in any domain that needs like my specific attention, then I would like just block everything else out and go very deep into that until it's solved and then move on to the next thing.

So, yes. And chatting to em, Maria, before this, you said you're very deep on multiple functions a across the company. Can we [00:52:00] talk about funding then as well, because that's something we've maybe allied to date or we haven't touched on. What was the initial process of getting funding? Talk to me about what that looked like, what you were looking for, those initial conversations, how things have developed over time.

So. I think in, in, in, in my case, because I understood that we're going into something that agriculture technology, it's a much longer cycle to crack than, let's say, a delivery app, right? And, and especially what we were doing in, in, in all three permutations of our products and business models. We had no playbook.

We were going, we were looking at a problem and we were trying to, from just design thinking and first principles, come up with the best solution possible to that problem and to problems that we hadn't seen anyone solve before. So I was looking for partners. That understood the nature [00:53:00] of the challenge that we're doing.

So I would care a lot about who the people are the fund as well in terms of who their LPs are, what their expectations are what's their deployment period. Are they at the beginning of their deployment or at the end, are they gonna start looking to, to, to make exits in the next two to three years?

Or are they comfortable waiting eight years to Yeah. To, to realize the return? And and, and, and. The main thing was people. So I've, we've, I think we've built a very unique but very high quality cap table. So with Amir from Tru, who is arguably the best early stage investor in the region. And Niten from Nuwa who is one of the smartest people I've ever met.

We had Foundation Ventures culture, a group of really, really solid angels as well came on top. And then most recently TX who were our customer from the get-go, like just before I even built the [00:54:00] app, when I was handing up flyers for farms to call us up for farm workers they were, they were kind enough from the very beginning to sort of like open up their farms for us to, to, to try solving that problem for them.

And over the years we've built a very good relationship and they ended up coming in and investing quite significantly in the business. So. I think what the, the number one thing was alignment in terms of that this is a mission that's worth pursuing and that it's gonna take time to, to solve for challenges in in agriculture.

Second, it's the people themselves that are people that I do want to partner up with and and, and, and work with and learn from. You mentioned time, it's often said that the fund life cycle is maybe incompatible with certain innovations , in agritech or the agriculture side of things.

Would you be inclined to agree with that? Is this very much you require patient capital? This is gonna take slightly longer. Yeah, yeah, yeah. Just because like farming [00:55:00] works with the cycles of nature. So on a good day if you, so first of all, you can never sell to a farmer during their season.

So you have to start talking to them in their off season. And if you do a good job at convincing them, they will try using your product in one corner of their farm and see if it works. And they will try it for a whole year. And then if it does work, then the following year they're gonna expand it a little bit and then the following year they're gonna grow, grow it to cover.

So to get a technology to be fully adopted on a farm, it's usually, let's say, three to five year cycle. Now, to do that at scale, right? You need to be incredibly inventive in your distribution and in your product packaging. And you need to have hit everything in terms of the value that the product provides.

The, the, the service, the after sales service that you provide, or the support service that you provide, the packaging of the product, how you price it. You need to, there's so many [00:56:00] things that you need to have figured out for all of these things to click for a farmer to then deploy your technology at scale.

And so I think that if you're, if you're, if you're really good, there's probably two to three years for you to perfect your, your recipe. And then you need three to five years to start seeing adoption at scale. That's in conventional technology and software. The playbook doesn't apply to AI anymore.

It's a very different playbook. But that's what typically, traditionally, traditionally would've been the, the, the timeline. How does the business model work? Is this usage? Is this per seat? Is this I'm curious about it too. Go on. , We haven't cracked it. So right now we're focused on value delivery which I believe we've cracked to some extent.

I think there we still have another good six months to really perfect it. [00:57:00] So I think by end of Q2 this year, we would've built an incredible, incredible product for the industry. I think already where it is today is very impressive. Even for me that doesn't happen often. But I think in six months time we will have something that is really incredible and that will add a lot of value to millions of, of farmers around the world.

How do we, how do we extract value in return for that is something we haven't, we haven't, I haven't even spent much time thinking about right now. Really right now. We just have like, in terms of like some, some usage limits where afterwards you need to pay $20 a month. Sure. So very basic. But that's definitely not gonna be how we continue to monetize out in the future.

How we do it exactly is still like, there's, there's different permutations and I think we just need to understand the value that we're giving to our users a little better. And also let the technology play itself out a little bit longer before we have a very solid position [00:58:00] on, on how to generate how we're gonna be generating revenues off this technology.

I don't think it's very, it's too far off. Like, I think we'll have a pretty solid idea in three months time and we'll start rolling it out in maybe a few, a few, a couple of months after that. So I think by, by mid this year we will, we will have had a very good understanding. But just today, I think that's not where our energy needs to be focused at.

How'd you get farmers to engage with it regularly? Is there any particular sort of re-engagement tactics? Question. You're on fire today. You're on fire. I think uh, excellent question. You, you, so the first thing is you need to just make the experience so frictionless that they're not coming despite.

the challenge of logging into a tool, you know what I mean? Yes, I do. I do. That's the first thing, like, don't treat yourself in the foot. That's kind of like rule number one. Just make it a very nice experience for them to do. And then the second thing is give as much unsolicited value [00:59:00] and what does that mean without the, the farmer even having to ask, give them value.

And that's core to our thesis and how we engage with os. And so orth today, every day we'll run a scan of every single one of your plots. It will look at the weather forecast, it will look at your field profile. It would look at what you've been discussing with it. Maybe a couple of days ago you were worried about some pest in, in, in infecting your fields or whatever.

Run an analysis every single day and it will send you a report telling you. What you might want to do, if there's any risks, it, so for example, humidity is gonna go up tomorrow at, at this time. So this might create an environment for fungal diseases to start to, to flourish. So you might want to apply some form of spray, or you might want to ventilate your greenhouse, or you might wanna do this or that to prevent or to decrease the risk of that disease popping up.

And, and so a farmer will just receive a notification [01:00:00] with, with all of this value delivered to them, without them even having to ask for it. And the way we, we, we see the future playing out is it will not only. And give you these reports, but it will be a core part of your system. It will be tied to your, to your machines, to your tractors, to your team members.

And in a couple of days we're releasing Earth teams, which is the collaborative which is, I'm kinda like low and sleep these days and talking a bit slowly. You're very coherent, you're very co ahead, don't worry. But and, and what we believe is that earth will become , the main intelligence layer , for farm.

And it would proactively assign tasks to people and to tell them, Hey, go, you need you. You might wanna go there and do X, Y, Z or, Hey, I'm missing a piece of information. Can you go to this spot, take a picture and upload it just for me to make sure that that this is that, that there's nothing fishy going on there.

So it's, it's really becoming this proactive companion . Whose [01:01:00] hands will Earth eventually be in then? Do you envisage a world where anyone working on the farm should have Earth as their companion? Oh, absolutely. Absolutely. That's, that's the goal from day one. One of the things that you touched on earlier that I'm just remembering now is connectivity.

. How does this work then, to your point, maybe starlink is improving this and we have a confluence of a couple of factors like AI and greater connectivity. . So starlink is actually improving this a lot in South America. Really? Okay. Yeah. So you'd be surprised how many firms use starlink in South America.

But I think that the main sort of like step change will come when you, when you start having a lot more of the inference happen on edge devices. Yes. So mobile phones are gonna be able to run small. Maybe at some point, large models just locally on the device. And I think that that's gonna be quite transformative.

And it's not just for farming, by the way. I mean this, this kind of like breakthrough. And I imagine like most companies are now investing a lot in that because for many reasons you just need to crack that for robotics. You need edge [01:02:00] compute to, to start , to be there. Machinery all of the, all of these things require the, they cannot operate in the physical world.

Having a dependence on being connected to a data center or something to run the inference, it's not gonna happen. So I, I think when, when this breakthrough also comes along it's gonna work. But I think that when this breakthrough comes along, I also doubt. That mobile phones will be the main modem of applying this kind of technology in the first place.

Okay. Well come on then. What, what, what, what, what do you think the new paradigm is going to be then? I don't know. It, it's, it's gonna be something that that merges the virtual world with the physical world, kind of like,, it, today, the closest permutation to that is like wearable glasses. Sure.

I'm not sure if that's gonna be what ends up evolving. You might just end up putting a contact lens and you have something sort of like somewhere and they communicate. And I don't know. It's honestly not something that I've sort of like dug into as deeply as some other things. [01:03:00] But I do think that that's gonna be sort of like the, because it, it just makes sense.

It just makes sense if you're able to run. A significant amount of compute on a very, very small device, and the use case for that compute is ingesting everything that you observe around you and giving you guidance on, on what to do to reach a certain optimal result. Then this seems like a much more practical way to solve for that than having a mobile phone.

How are you using the funding? So we're using the funding, so it's, it's 70% is going into r and d . Product development. And then the rest is, so 20% is going to marketing and commercial. But that's shifting. So I think by, by mid this year we'll be at almost 45, 40 5% split, and then you have 10% on like overheads and admin and and, and [01:04:00] legal and stuff.

Is the workforce management is this, that's still an offering. Yeah. That you have. Yeah, yeah, yeah, yeah. And I'd imagine, does that account for a greater revenue percentage at the moment? At the moment, yes. In six months time it'll not be the case. No. , Someone clicks on an ad for, for Earth, how easy is it for them to get up and running on the platform?

Time to value is three minutes. . So from the moment they download the app till the moment they've extracted the first value moment. So they've downloaded the app, they've signed up, they've been onboarded, and they've received some form of value or prescription from Earth. The average now is three minutes.

Just something you've really optimized for and refined. Yeah, we started, we, we started with it being about 27 minutes, and in the past couple of months we brought it down to three minutes. And now we're thinking of prolonging it a little bit to maybe four or five minutes, but in exchange of greater value at the fifth minute.

So it's kind of like the trade-offs that we're thinking about right now. Explain that rationale to me. So there's, there's always [01:05:00] sort of like a, a, a trade-off almost in everything in life between quality and speed, right? And so you can optimize for speed a lot, but then it can decrease the quality. And then you can go a bit slower, but then deliver higher quality.

And at the end of the day, it's a spectrum or it's a scale that you can, that you can manage. So right now we've optimized for speed. Not to say that you don't get quality, you do get quality, you do get a quality response. But there is a, there is a premise for getting even greater quality if your onboarding is slightly longer and you're, you're sort of like engaging with earth, giving him a little bit more context at the beginning that might ultimately yield to but these are all experiments that will be running in perpetuity.

Yeah. I think generally when, when we think about system design and that's why I, I said like with ai, the playbook is completely different because with ai, the fundamental premise is you need to build a system that does not rely on having any humans in the loop. [01:06:00] And you can, that's what's new.

Before you, people could say that all they want, but to execute on that, it was virtually impossible. But today you actually can, you can have someone come in, start using your product grow with the product and you can have agents taking care of everything around it. Not to say that also that's fully possible today.

Like if, if in some places if you like, need a document stamped by bank, you have to go to the branch and you have to do it. So you're limited by what your environment gives you. But the direction of travel, the direction of travel, pretty evident. Exactly, yes. Right. And so we, we've, we're building earth from the get-go with that vision of the future embedded into the business model where we are trying not to have any humans in the loop at any point.

And so up until today, all of our users and all of our subscribers that have come in through had zero human interaction with [01:07:00] anyone. Even with support, if they have a support query or to my customer or whatever we, we, we, any support query that that comes across. So we've sort of like have an email where people can send, we have Sure.

Once there's a support query we update the knowledge bank for Earth so that it can give the users responses to any of their questions just in the chat. And it's, and I think that that's gonna continue to evolve. So now we're thinking about, as we're launching teams in a, in a couple of weeks, it's gonna allow businesses to collaborate and to have an account.

It's the same, no humans in the loop. And, and then for enterprise, this is where it starts getting a little bit trickier which we're planning for in, in, in the second half of of this year. Because enterprises, I mean, they just need some form of human involvement. And going back to what I was saying earlier, the value with enterprises is not gonna be so much that you will need a human to configure the system and set it up we'll make sure that this is handled purely.

Through the technology. [01:08:00] But when I was talking about remote work versus being physically present, this is where humans are gonna add the flavor, right? This is where having someone from earth, the, the earth team or organization, sit with your team, work with them, get them to adopt and to get the most value possible from the product and so on.

I think this is where humans will, will, will be able to continue delivering a lot of value in the future. Something I've been thinking about recently is , one of the constraints today is that. When I go onto Excel, for example, the interface that I see is the same as the interface that a chief financial officer of a company sees.

We probably do not use Excel in the same way. It's probably fair to say it doesn't really make sense that they're both the same. I suppose with the advent of AI now there's maybe the possibility that that can be more customized or tailored to my particular use case and how I use the product. We both don't need the same level of details or tooling.

Do you envisage the [01:09:00] world with Earth, for example, at some point that you know, someone in the back office using orth, a slightly different interface or , they don't need to see the exact same things as someone out in the field? I think it's a very, so in the, in the, in the example of Excel, I think Excel has been so successful exactly.

For the reason that you state as an inconvenience. Mm. I think if, because there's been many attempts to replace Excel for specific use cases, they never succeed. This is true. And so the human brain is conditioned to like absorb simplicity much better than complexity. And, and

I see a case for for the software having different versions when communicating with other AI agents. So I think that AI agents are gonna consume software in a very different way than humans do. So today what's happening is, like, for example, Claude just [01:10:00] released their Excel agent, right? That sort of like works on top of Excel to build the models and stuff.

And this is purely because. The output is designed for a human to consume. But if, if, if Claude is building a some form of a financial model that's gonna be consumed by another AI agent and that no human will ever see, I doubt that it would build it on Excel. It's a really good point. I've actually talked to Aria about Shannon from applied ai, AI about this previously.

And to his point, they were expecting the agent to work through a problem in the way that a human works through the problem. And they're really perplexed as to, okay, the answer is the exact same, but it's kind of about an entirely abstract way. Then we would be used to, but it doesn't matter. Yeah, because as, as humans, we, we optimize for the presentation of the information so that it's consumed in the most.

Easy, digestible, digestible way. That's why like if you're a top tier executive, like you will say, you'll tell, just present to me in [01:11:00] one slide. One slide. Send me two bullet points. And you know, so you really optimize for, for that. And then you rely on all the context. You have to make sense of what's in front of you and to place your judgment and storytelling, I think humans also optimize a lot for, for storytelling and for consuming engaging stories.

Agents , will ultimately optimize for token consumption. And, I think it's a very different approach. And so they don't need to have all of these visual elements. They don't need to build all of these visual representations and slides and make things look pretty and nice and stuff.

They just need the, the bits get to zero to one A to B as efficiently and as efficiently as possible to deliver the the outcome. So I think that that's where there's gonna be a very different kind of like consumption pattern for, for, for software with agents. And I have no idea what that's gonna look like.

But I think it will reveal itself in due course. But I do think that humans will [01:12:00] more often than not, opt for a uniform experience that is harmonized across different people. 'cause there's a lot of advantages to that. Yes. You can learn from people who are using it. So they will upload videos, use cases, your buddy, your neighbor's using something, he'll show you what he did with it.

And so you don't have to. You don't have to put as much effort in learning something and in discovering it, you can just absorb what other people are doing and this kind of like knowledge compound. And so I don't really believe in the idea that every single organization is gonna go out and build its own set of software tools for every single thing you will have use cases where Yes, makes sense.

They, it makes sense and they will do it. And maybe someone who's doing something that's very narrow and wants to like digitize a certain process that doesn't exist or that's very expensive. Otherwise, yes, they will do that. But for the most part, like to replace an ERP, for example, or a core system of record in a business, I don't think just get rid of Salesforce.

We'll just vibe code it. [01:13:00] Yeah. That's, I, I don't, I don't think that that's gonna be the case and not because technology is not gonna be able to do it. Because human beings thrive on standardization and sort of like common, like having a common pattern in terms of usage. And, and these then compound the benefits.

So if you're using Salesforce, odds are your auditor yes. Will know how to read your books and prepare the financial statements versus and, and even a tax authority will come in and they will consider that you're keeping your books properly in order and they'll be able to go through it and, and, and, and work with that tool.

Versus if you have some vibe coded accounting system that might not meet all the standards, that might not, that's not certified in any way. Certification is a big thing as well still in our world. Even onboarding transferability of just employees in terms of adaptability. It's such a good point. Are you having fun?

I'm having a lot of fun. Yeah. That's the most important question [01:14:00] I was. I was in the, the office 700 last week and I was like, yeah, that was, I was just setting the team like this is so much fun. And especially when you're, when you're, when you're building something that, that resonates with you, that has a lot of meaning and at the, and, and you know, it's very useful.

And at the same time it's at the leading edge of technology. So like, I'm, I'm naturally like a very curious person and so my, my brain is like constantly on overdrive now which is which is a lot of fun. And then you solve different pieces and you, you start seeing things work and yeah, it's. It's endless possibilities.

And, and, and you get, you get to hang out with farmers in beautiful places all around the world. You get, you get to like, there's this whole other aspect of my job besides like the whole technology piece, which is most of my job actually. It's like spending time out on the field talking to farmers. Yeah.

You're fluent in Spanish now. Like, so I mean that's a byproduct nearly in many aspects. I learned, I learned Spanish a couple of years ago when we set up shop in Madrid. 'cause I was spending most of my [01:15:00] time at. In the southern coast of Spain, around and and Seville and it's a tough life.

And these places, Seve's a great city. Syria is stunning. Cordoba is beautiful. I went there when I was a lowly student, so I think the only place I used to eat was like, you know, Montoss, the little kind of like sandwiches. Yeah, yeah. That was my uh, s and then like a cerveza, you know, you're, you're flying.

That's all you need. That's pretty much it, man. That's very much it. No, and so, so yeah, so I mean, I, I learned Spanish and it's, it's a beautiful language. It, I guess it was easier for me to pick up because I, I, I speak French. Oh, okay. So it's kind of quite similar, but still I suppose this lends itself well to your curiosity or sort of an, a curiosity.

Can you talk about discipline then is something I'm, I'm interested in. Like how do you restrain yourself from, 'cause I find this myself because. Rabbit hole, like, that's so interesting. But to, to your point, you're still trying to, you know, there's a, a roadmap I'm sure that you have in place internally and you're trying to stick to that.

How do [01:16:00] you No, I, I try to stick to the truth. I guess that's one way to put it. So I have, I have a roadmap where the only thing that can change it is if, like, the truth reveals itself to point to an opposite direction. That sounds mystical. Not like, like you can have a roadmap based on a certain set of assumptions, and then if you're, if, if one of the assumptions changes, then you have to change, right.

Otherwise you're, you're just being stubborn. So there's, there's a difference between the two. I, I generally I think it's, it's, it's in the early days of starting the business, I think there was a lot more curiosity than discipline. I think today it's quite. The opposite. Sure. And so I have a, like a, I have a calendar that I follow religiously and like every single week it's almost a repeat of the same calendar that I've optimized and learned over time.

Like, it wasn't kind of like a anything. And so I do have time allocated where I can just go into a rabbit hole about something [01:17:00] and, and, and learn something. We have to, is it Google or Meta do that? What's it, the 20% project or something that, I dunno what exactly they call it, but they allow you a certain .

Like day in the week Yeah. Where you can work on your own for, for me. So, so I work I work around 14, 16 hours from Sunday to Friday. So Saturday is the one day that I switch off. And and Friday is the day where it's mostly for my own, curiosity, pursuit, sort of like research and stuff, and experimentation.

And then every other day is focused on a specific aspect of, of the business. Interesting. How, how do you divide your time? I'm curious about this, like, is it a different function that you're focusing on on different days? So there's some elements that are function specific, but then really what, what guides about 60 to 70% of my time during the week is what we identify as the core bottleneck in our weekly business review.

So we have a meeting every Monday, which is called the Weekly. [01:18:00] So Sunday is for product, just because I like to think about products with no restrictions on the figures. On the numbers. I don't like the business metrics to influence my product thinking. Yes, yes. Right. And the, then there's Monday morning we have the bus business review where I look at the metrics and we see where we have a bottleneck.

And that bottleneck is basically what I focus on until it's solved. And then there's other like, check-ins that are carved out. So I'll check out, check in with our marketing and commercial. I'll check in with my chief of staff who looks at hr, finance, legal, all of these things. I'll check in with our customer success.

But unless un, unless they are involved in the subject of that is sort of like the core priority, then it's just a check-in just to make sure there's no fires. But I try to go very [01:19:00] deep on one, one or two things per week. Has your leadership style changed over time? A lot. Yeah. Yeah. How so?

Significantly? No, I was, I was I was insufferable. I guess

I've heard that myself, so don't worry. Yeah. I, I think I, I, I sometimes could still be but for the most part, I think I've, I've understood the value of thinking in terms of systems instead of thinking in terms of individuals or or acts of heroism. Mm-hmm. So at the beginning it was a lot of acts of heroism.

, And now it's a lot more system thinking and system design. And, and focusing on how the system runs. How has that evolution come to pass? Is that just something that took time? Has it just been a gradual evolution and progression? Yeah. I think it's just been a gradual thing. I, I, I. It's, it's the same as like, how am I think if your, if your end [01:20:00] goal is a success, then then you should never, like, the same way you're curious about problem solving and technology and product or whatever the same level of thought and self-awareness should apply to your leadership and your management.

And, and, and these things I think should come hand in hand if, if you're really, if you're really gonna be successful. 'cause without that, you're not gonna be successful. Yeah. It seems strange to optimize all other facets and and leadership to be an afterthought doesn't really make much sense. Yeah, because at the end, at the end of the day, like you can, like if, if you're going at it alone, there's, you can only go so far.

Versus like when you're operating with a highly motivated and highly capable group of people. You, that's when you can really achieve some, some incredible results. Any interesting Friday experiments? The last couple of weeks? I think this, this one's quite technical in our field.

But essentially there was, we were [01:21:00] facing some form of a challenge in terms of like disease identification. Mm-hmm. And so I, I tried playing with different things. I tried like using some existing models that machine learning models that were specifically built for disease identification.

They're very difficult to integrate with LLMs. And they're often sometimes a little bit narrow as well, focused on a specific crop. So they, they don't fit with our principle of being very broad and very like successful. And, and the LLMs themselves are not built for that. So the. Imagery, sort of like detection and, and, and pattern detection to identify disease accurately wasn't doing so well.

So I've, I mean, I've tried a whole bunch of things and, and then realized that because it's probabilistic, and so the more input you provide it, the higher the probability that it leads to the right outcome. And so if you, I mean, it's quite, it [01:22:00] sounds quite simple, but if you just give it one picture, picture, right the, the odds that it gets to the right outcome let's say is 60%.

If you give it two pictures, the odds of success are compounded if you give it three pictures. And so when you get to four pictures. The odds of success are almost at 95%. That's pretty increasing that surface area. Yeah. And, and you give it different pictures from different angles or one of the leaf, another of the the tree and, and another like check the so and, and so, and, and, and when I thought about it, that's, that's how humans go about it.

Yes. Like if you probably, if you, if you send someone one picture they'll tell you absolutely it could be this, it could be that, can you show me this? Can you show me that? And you sort of like eliminate by process of elimination, you get to the right answer. And so you're able to sort of like carve this as, as, as [01:23:00] a hack in the, in the workflow without using any other models.

And, and, and the accuracy is gone up quite, quite a bit. That's so interesting. Yeah. It's just a difference in, in where it's dif different in thinking just approaching it, it's not actually anything wrong necessarily with the underlying and No, no, no. And, and, and that's the thing. Ai, you need to think about it not as a technology.

It's you're dealing with almost as if you're dealing with a person. If I, and, and going back to the, when I told you like the environment and then the thinking and then the output. If you if you hire someone in your business, what's the first thing you do? You onboard them. You give them a laptop, you're like, here's your tools.

You introduce them to their ma, to their managers, their colleagues. You give them the company policies. These are the dos, these are the don'ts. These are your KPIs. This is your contract. And you give them time to get onboarded. Right. To learn. Mm-hmm. Right. And then throughout [01:24:00] their work, you teach them how to think.

How to operate, how to behave. And then at the very beginning, if it's a fresh grad, odds are you're not gonna let them go and present to the CEO o of the, of the company. Right. But over time, as they get more confident mm-hmm. In what they're building and and in what they're doing, and in their ability to explain and to re and to explain their reasoning and present it in a proper way, then you start letting them present to small audience than a bigger audience.

Than a bigger audience. Right. And that's the natural process that happens with humans. Right. And so that's the same way that I, I approach AI philosophically, you know, like that's the exact same process that I try to apply when I'm, I'm dealing with with, with agents. The only difference is to go from a fresh grad.

To an employee or to someone who's gonna present to a board or to a CEO or whatever. [01:25:00] If you're really good, it's gonna take you five, six years. Like if you're exceptionally talented in your, in your, in your field with ai, you can today with where it is today provided you're giving, you have the right environment ready and provided you know what kind of outcome you want from it, you can get it to a pretty good place in a couple of weeks.

Yeah. Which is just insane. No, it's an amazing analogy and I think , it's something that I've thought about from a consumer adoption perspective. I think that's the great lock in that chat GPT has, is how much it knows about you, , the amount of information that you've been feeding it over time, the level of personalization, and that's maybe.

I certainly was reluctant to engage with, you know, when I was using Claude for the first time, I was like, it doesn't really get me, you know, this is, this is frustrating. Yeah. But all I took was, you know, I used, Claude was like, I'm just gonna use Claude for a week, and just going to, you know, feed it as much context about who I am, what I do, the way I think, the way I prefer answers, et cetera.

And then there was no going back. But obviously if you provides very, [01:26:00] very little context then, so you use, you only use one of them now at a time? I, I go between all three. I use chat pt. . Less and less though. . Claude predominantly. . And generally I've, I found myself, which is really strange in many respects, using Gemini for things that I historically would've Googled.

. Because I find that it's better t trawling the internet, I find, in terms of how it presents facts, it clawed very much for workflow. . And chat. GPTI still occasionally use for, I think it gets my tone. Going back to the analogy about AI and, and human beings, no human being in their right mind would ever say, I only talk to one person.

It's true. It's true. . Right? And that's gonna be the same with ai. You're gonna be talking to dozens of agents. The only thing is these agents are like general purpose and they don't explicitly say, this is what we're really good at. Yeah, use us for this. [01:27:00] They try to like, because they're after market share and, and use cases and stuff.

But I think that over time each person will know which agent to go to for what task. It just becomes intuitive and just becomes intuitive. And you will have also agents that will do that on your behalf and be kind of like your assistant. But yeah, I think what you're saying is absolutely right.

And then you'll start. Disliking some of the agents. Mm-hmm. Almost in terms of a personality, like when you meet someone and you don't like them, you'll start developing this based on how they write, how they respond, their tone. That's what's happened with ChatGPT. It started giving onto slightly differently.

Totally. And I went, I do not want to, I don't wanna talk to you. Yeah. So like subconsciously you start developing a relationship with them and it's creepy. It is. Despite your best efforts, because I, I consider, you know, I can be an attached objective observer, but you, but be, but because they communicate and they react to your actions.

Yes. And this is kind of like the core foundation of any human relationship. I say, [01:28:00] if I tell you hello, I will say hi back. You'll say hi back. Right. And so that's the start of a human connection. And, and so if you tell an agent hi, if you go to, if I tell this microphone hello, it's not gonna respond to me.

I'm not gonna develop a relationship with it. But there is something is communicating with me. I will, if, if it cracks a joke. To me, I, I, I, I, I like it if it writes in an aggressive way or it's, I feel that stone is condescending or any of these things, I'll, I'm not, I'm, I'm gonna dislike it. And so a big a and, and, and it's, I don't think that AI can sort of like self govern itself, at least yet in that aspect.

And this is where having a human touch matters, matters a lot in term. And, and it's going to be a significant contributor to enterprise value for any AI company is going to be the, the, the, the, the tone and the, the character of your agent [01:29:00] as well as the, the design. And the brand that's built around it.

And so for us, this is one of the top priorities. It's, it's the brand and the, the, the design language that, that we use for the product. It's beautiful. Stunning. Or we have, we're a very young company. We're very like small people, but we have a head of brand who's shout out Kareem. Kareem. He does an incredible guy.

He was, he's who's a creative genius and, and I take that very seriously. It's really unique that, that was what I would say that's one of the most striking things initially when I came across that. Because in, at a, at a time when at, in an, in an age where knowledge becomes commoditized taste.

Yeah. Claude have done an amazing job of this. I think personally, I think initially didn't exert any influence, but over time I was like, Claude's the cool model. Yeah. Claude is the model that creatives are using or that, you know, chat. GPT is the masses. You know, Claude is the more unique sort of detached, cool hipster model.

I wanna be associated with that. A hundred percent. A hundred. Yeah. And that's gonna be very important because as I [01:30:00] said, like humans, they, they, they, they get impacted by storytelling, they get impacted by beauty, the presentation of things. And so assuming that the functional element is solved for this is what's gonna set you apart.

Okay. Hassan, they're gonna try us outta here, so we better finish up. Is there any question I haven't asked you that I should have asked you? You've, you've asked a lot to be honest. I don't know. I mean, there's a lot. We'll, we'll take that more. We'll take that. I've done an amazing job. Yeah. No, this, this has been a lot of fun. Yeah, this fun. Thanks a lot for, for hosting. No, thanks so much for coming on, dude. Yeah, absolutely.