You’re Not Thinking Big Enough: Rebuilding Food Systems with AI Agents — Cody Menefee, Firecrawl

0:01 [music] All right. Hello everybody. Uh my name is Cody. Um I put this picture up here because this jacket so far has not actually landed as well as I thought it

0:23 would. No one gets the joke. Um so this was to really put it in front of your face. I don't just enjoy wearing heavily branded Letterman jackets. We intentionally tried to play into the bit

0:34 a little bit. Um, so my name is Cody. Uh, I'm on the growth team at Firecrawl. Um, and today I'm here to tell you you're not thinking big enough. Um, but before I actually can get into that, I

0:45 have to address a bit of an elephant in the room, which is Theo stole my talk. Uh, Theo put out a video about a month ago uh called uh, you need to think bigger. But I would like to say I

0:55 submitted the name for this talk a month before Theo put his video out. I didn't steal his talk. He stole my talk. So, um, today we're actually going to talk about farming. And yes, I actually mean

1:06 farming. Uh, more specifically, I mean livestock farming. And even more specifically, I mean automating pasture rotation for grass-fed livestock systems. Um, I have a feeling most of

1:18 you did not expect to learn about cows and grass and farming today, but I'm here. So, you're going to uh little bit of background on who I am and why maybe you should listen to me. Uh the short

1:30 version is I actually have no credentials that qualify me for this talk, but nonetheless, I'm going to do my best to give it. Uh I grew up in Kentucky. Uh have a background in blue

1:38 collar work. I was a bartender, a mechanic, a uh um server, uh a whole bunch of things. Uh never actually a farmer though. Um and then I sort of found my way into engineering, uh

1:50 software development, etc. I actually don't really like taking the title of software engineer. Um, I'm pretty adverse to that. Feels like stolen valor because I am the vibe coder most of you

2:00 all are scared of. Um, I use AI agents all day long. I don't have any syntax memorized. I am not proficient in any particular coding language, but I will crank out some stuff on a weekend. Um,

2:11 but today I actually work at Firecrawl where we're building context for AI agents. Uh, we have a series of uh web data APIs. So, you give your agents access to the web. Again, I said I'm on

2:21 the growth team. Um, but today we're actually to get into some more farming stuff. Uh, but first, a bit of credential I do have is this is a real picture of me hauling turkeys on top of

2:31 my Tesla and I do still have a crack in that glass ceiling because of it. Um, this was in Nashville where I live in the middle of a residential neighborhood where you are not allowed to raise

2:41 turkeys. Um, but I raised 10 turkeys in my backyard because I wanted to know what it was like to actually uh raise livestock myself that I would eat. Um, it's a very mentally

2:52 uh difficult process if totally honest. Um, but this was me loading them up onto the roof of my Tesla and then I drove for three hours with them to the processor. Had to stop at a supercharger

3:02 on the way. Um, and lots of people were taking pictures and what I can tell you is is your range is sufficiently decreased when you have a giant windbreak full of turkeys on top of the

3:12 roof. Uh, so that was uh quite the anxious drive, I can tell you. Um, I also almost left uh engineering to be a farmer. I really wanted to raise chickens. This is a real product image

3:24 that I came up with. I wanted to wrap turkeys in white wrapping and literally just slap the word eat me on top of it. I thought it was provocative. I thought I would get you to buy chickens. Um, but

3:34 I realized it's actually really hard to make any money farming. Um, surprise surprise. Um, and I have a wife. I have two kids. it didn't feel uh right to ask them to give up the lives they had so

3:47 that I could go cosplay as a farmer and raise chickens. So I decided to pivot and see if there were ways that we could scale farming itself and the the types of systems that I'm interested in when

3:59 it comes to livestock agriculture. So three things I want to get accomplish in this talk is one convince you all to pursue bigger ideas. Um I think a lot of these talks, a lot of these conferences,

4:09 a lot of us individually uh uh spend a lot of time talking about building software for people who build software for people who build software so on and so forth. And I really am just here to

4:18 challenge you that there are other problems to solve than just another MCP for another SAS solution at another company. Um but also I'm just really trying to take advantage of a captive

4:28 audience. Uh if you corner me anywhere at any time, there's a good chance I will talk to you about farming. Um so here I am. Uh, and hopefully I can convince you to come work at Firecrawl.

4:39 So, uh, first things first, I believe livestock belongs on pasture. I think animals should live on grass. Um, I think it's better for the animal, the consumer, the farmer, the ecosystem. I

4:48 can give you a whole TED talk on each of those if I need to. You can find me later if you need me to tell you why it's better for animals to be on grass, but I don't have enough time to get into

4:56 all of that. Take my word for it. Let's start there. The assumption is animals should be on grass. Um, this is the goal I want to hit. I am not actually anti-containment farming. I

5:08 think there's a reason we needed to do that. But 97% of cows are still currently pro uh finished on feed lots. 3% are raised on pasture. My opinion here, more animals could be on grass. I

5:18 want to try to figure out how we get more animals on grass. The question is why aren't they on grass? And that is labor is the bottleneck. It is a pain in the ass to actually raise animals on

5:27 grass. Uh pasture done right actually means moving animals constantly and that takes a lot of work. Um if you think about grass-fed uh beef, you might think of I have 100 cows, 100 acres, I put 100

5:39 cows on 100 acres, they eat grass, I got beef at the end of the year. That's not quite how it works. Um you will uh very rapidly decrease the quality of your pasture if you just let cows graze where

5:50 they want because they'll graze their favorite things, ignore things that they shouldn't, trample areas consistently, so on and so forth. So, the solution to that is rotational grazing. What this

6:00 means is you break up your pasture into individual paddics where the animals have enough uh food for one one day and then you move them every single day. Um, this allows certain areas to rest and

6:11 other areas to be grazed and over time will increase the efficacy of your pasture. But this takes a whole whole lot of work. Um, this means you have to move fences, animals, water, and keep

6:24 track of it every single day in order to uh appropriately move the animals as often as they need to. There are some solutions actually trying to work on this problem. Uh, you may have seen a

6:37 company called Halter uh in the news recently. Peter Teal invested at a $2 billion valuation. Uh, No Fence is another company. What these companies do is provide collars for the animals

6:48 connected to GPS satellites that allow you to draw virtual boundaries where you can move the animals uh remotely. Um I think this is a great step in the direction of trying to uh expand labor.

7:00 Um but this has a problem which is you have to know where to move the animals. This is not a science to actually be honest with you. you can't just move them in a straight line across the

7:11 pasture routinely every single day to the same part of land. Um the reason is is grass doesn't grow the same every single day. Uh there are drought conditions, rainfall, uh how much impact

7:22 a particular section of the paddic has had. And the way that this is sold today is actually farmers going out on pasture putting eyeballs on the grass and making intuitive decisions about where the next

7:32 best move should be. So the question is, how do we replace the farmer's eyes on pasture so that they can remotely make educated decisions on where to move their virtual fences?

7:46 Uh there's a bit more that actually goes into this as well and that is you can't just you have to also know how tall the grass is. Um grass has a growing cycle. If you grow it way too short, it takes a

7:56 really long time to come back. If you let it go too long, it becomes old and bitter and the animals don't like it. There's this juvenile sweet spot that you want to keep the grass in. You want

8:06 to cut it before it gets too tall, but then you also don't want to cut it too short. You need to keep the animals moving and then constantly coming back to the same pasture so that your grass

8:15 stays at the most optimal growing age and constantly has uh the most productivity possible. So, a couple of ways that we can do this. Um these are things these are my solutions. This is

8:26 something I've actually been working on thinking about how we can do this. Um, a couple of options that I have are drone orthamosaic maps. If we could find a way to automate drone flights, we could go

8:36 fly them around our pasture, take a whole bunch of pictures, get some very highfidelity, high resolution images of the graphs that farmers could analyze. The problem with this is uh there's a

8:45 lot of uh skill upgrade you need to do with the farmers to teach them how to fly drones. A lot of regulatory issues with keeping the drones in sight. And ideally, this would be autonomous. And

8:54 there currently isn't a jurisdiction in the world that has approved autonomous drones for these types of applications. So this is a really big bottleneck. Um I think it has pretty high fidelity in the

9:03 quality of imagery, but is going to be a hard problem to solve in terms of actually getting all those hurdles accomplished. Uh satellites is my most favorite option today. There's a really

9:12 cool company called Planet out there taking pictures of the entire globe every single day um with a uh one by one meter resolution. Um, but there's still those satellites are really high up in

9:25 the sky and uh it's hard to tell some of the things you need to tell uh to actually make those educated decisions. Uh, the middle photo here is actually from a friend of mine out in Missouri

9:34 working as a research grad assistant at the Missouri Lincoln University. And this idea is just putting a trail cam next to a tree and some measuring apparatus that that camera can look at

9:46 and just figuring out how tall is the grass in relation to that particular object. Um just so that we have some sort of reference point that we can use to see how well the grass is growing

9:55 back. If we can solve this problem along with the um uh the uh the collar uh situation, I think there's a world here where we can drop an LLM in the middle of this loop and start to work on

10:08 autonomous grazing operations. And so what this would mean is an LLM essentially making the next best decision on where the animal should be any given day. And this is a multivaried

10:20 analysis. This requires yellow to have several data inputs including where the animals are in GPS location, where they were yesterday, where they might go tomorrow, what the drought condition is

10:30 in the area, how tall the grass is across the entire farm. Um, and actually it has to make this decision not just on day-to-day basis, but in varying uh degrees of relation. So where's the best

10:43 next place for a particular cow to be, but where's the best best next place for the herd to be in relationship to the pasture itself, in relationship to the farm as a whole, and then more broadly

10:52 the ecosystem at large. There's uh all of these components feed back into each other. And if you can optimize this entire picture, you have a more productive farm uh where you can

11:02 actually have more animals on fewer acres. Um which is how we end up actually scaling to compete with the feed lot uh style where you can actually have more cows on fewer grass.

11:12 How do we solve this problem? Um, there's a couple of components. There's three main blockers that I think need to exist in order for us to actually create this system. The first one is building a

11:21 knowledge base. And this is primarily what I'm working on at Firecrawl. And then an open source project I have called Open Pasture. The idea here is a lot of the knowledge on when to move,

11:30 why to move, how to move, uh, uh, the benefits of moving, etc. is all locked up in primarily YouTube videos. There's a bunch of really cool farmers out there. I can give you a whole bunch of

11:41 channels that you can go down rabbit holes on of just good old guys out in Missouri, Tennessee, Kentucky trying to move their animals every single day telling you what they're learning,

11:50 telling you what species are best for this, uh what legumes are you want to uh aim for in the biodiversity and your pasture. There's a whole bunch of things that go into this and we need to build

11:59 that knowledge base. Uh Firecrawl is a toolkit that I use to actually collect this data. um going out scraping those YouTube videos, scraping research papers out to archive, building this knowledge

12:10 base up and Open pasture is the actual repository I put this information in to make it available to any farmer I think that might be able to use it. The next thing to solve is the actual

12:21 visualization layer. There's a lot of components we need to know about the grass that the farmer is primarily getting out of the intuition from looking at the pasture. The two main

12:30 things worth figuring out about the pasture, both where the animals are, where they should go, and where you want them to be, is what is the biomass, how much foliage actually is available for

12:40 them to consume. And then long term, what is the biodiversity of that particular pasture? Uh if they overg graze sections too heavily, they'll start to overindex on uh different types

12:51 of cool season, warm season grasses, legumes, etc. And ideally, you want a really rounded, really diverse pasture over time to make sure that the cattle are getting the nutrients they need. So

13:01 you don't have to supplement with things like hay, copper, aluminum, etc. Um, ideally they get all of the uh macronutrients and micronutrients from the grass itself, which becomes an

13:11 entirely ideally hands-off system. And then the third one is uh those geoence uh companies. So, no offense, Halter. While I appreciate the technology they're trying to push

13:22 forward, I have a pretty strong disagreement with them, which is in order to use their software, they require you by their callers, and you can't plug your own software into their

13:31 callers. From a business standpoint, I get why this is. From an industry standpoint, I think it's really a pain in the ass. Um, I would like to innovate on the software layer. I would like to

13:40 push GPS locations to these callers that my LM can predict. I don't want to have to rely on their software to do this because I don't think it's as good or I think I could make it better. I'm being

13:50 totally honest with you. Um, so a bit of the purpose of this talk is actually a call to action for you all in the audience. I need someone to make me a caller. Um, I need it to be open. The

14:02 APIs need to be open. Ideally, it's an off-the-shelf solution. Uh, some component parts that we can slap together. Uh, farmers are pretty scrappy and like to heal their own things.

14:11 there's a lot of uh animosity towards John Deere and this sort of like right to repair. Um so my ask to anyone maybe looking at this problem is design me a caller where the patent can be open and

14:22 the APIs are open so that we can compete on software and optimize this solution. Uh the next thing I'd like to maybe tease you about is this actually goes beyond just ruminants. Uh so ruminants

14:33 is a type of animal. Cows, sheep, goats, uh those are all ruminant animals. They chew grass. they digested in the rumin which is an organ called ruminants. Um but there's actually an additional

14:44 benefit we get where we can stack species on these pastured rotations. This is a company called pasture bird. Uh they were a big catalyst for me to really getting obsessed with this idea.

14:54 What they did is took your normal chicken house, put it up on big wheels and automated the movement so it creeps its width every 24 hours across pasture. The reason you can do it this

15:05 scientifically with chickens is because they don't actually get most of their nutrients from the grass. You have to supplement them with grain feed because chickens are omnivores, not quite just

15:13 herbivores. Um, so you can just inch this coupe ac across the grass. Uh, giving them uh uh fertile land to grow on. Uh, the nitrogen from their droppings actually help uh as a manure

15:27 uh or as a as a fertilizer for for the the grass itself. And there's an added benefit here when you stack the ruminants uh with the chickens. If you run your ruminants first, they sort of

15:35 cut off the top of the grass um and the chickens come behind them and peck out the parasites from their droppings and it reduces the parasite load overall across your farm, which reduces the

15:45 medication expense you have to actually pay to keep your animals healthy. And over term, you have a more robust uh uh seed stock or uh breeding stock so that you can have stronger animals over time

15:55 that require fewer interventions and can be left alone to just eat grass. and uh turn into meat eventually. Um so the three things I really hope you all take away from this talk is one I

16:07 think we need to find big real world physical problems that we can solve um that require multivaried analysis and not quite uh yes or no decisions. Um there are lots of problems out there

16:19 that don't actually have deterministic solutions. I hear a lot of engineers talk about how we turn LLMs into deterministic processes and my contention is actually there's a lot of

16:28 problems that you can't solve with deterministic uh algorithms. Um this is one of them. It's a multivaried analysis. There isn't a next best paddic to move to. There's just your best guess

16:39 on where you think they should go. Um, and I think if we can take systems like that, these uh multi-data input systems, and drop an LLM in the center to actually reason over the data and at

16:49 least make a suggestion that the human can confirm or deny, um, we can really start to scale systems like this that are very much, uh, restricted by the farmer's ability to scale their own

17:00 labor, their own decision-making power, um, and really give them the tools that they need to grow their operations to hopefully, I think, uh, all animals could be raised on grass if we, uh,

17:10 solve these problems. And then the last one is is maybe you can come help me solve some of these problems. Um the number one problem is actually just giving the agents the context they need.

17:19 Gathering the data, packaging that data, and then presenting it in a way that the LLM can reason over. And that's what we do over at Firecrawl. Uh so you're not thinking big enough. Firecrawl is where

17:30 we're building the context layer for AI. And I hope you can come build it with us. Uh we're hiring. Uh so here's all the job postings we currently have. go to our website, maybe find one that

17:39 works out for you. Uh, reach out to me. We'd love to have more people trying to figure out how we get data off the web to solve some more of these complex problems and present that that data as

17:47 context uh to these AI agents. And perhaps we could make the world a better place. My name's Cody. Uh, Open Pasture is my open source project. Firecrawl is where

17:58 I do my day-to-day life. And, uh, these are my socials. I'll hang out for a little bit. I'd love to chat more about animals, cows, birds, all the like. Thank you very much for coming.

18:23 >> [music]