It was the first day—not long after we started cutting— and we had a draper belt malfunction. Fun. The draper belt is the rubber conveyor that carries the soybean stalk towards the throat of the combine so the machine can eventually separate the seed from the stalk. And this particular situation wasn’t a “duck it or F&$# it” moment. The belt was ripped in half— it had to be replaced.

But no worries. We farmers actually expect things to break. When things don’t break, we expect a meteor to fly out of the sky and land exactly on top of the only piece of equipment that hasn’t broken down in the last week. It runs in our blood.

Luckily, the local CaseIH dealer had the draper belt I needed so I made the quick trip to Clarksdale to grab the part. In about an hour and a half, the machine was back running. But the story begins there.

Am I the only farmer on planet earth who finds it borderline impossible to stay organized during the busy seasons of life?

The story of the draper belt has historically ended once the combine is back up and running. The bill gets paid, the part is expensed, and the beans get cut. But what about that invoice?

I know, $1521.50 for a rubber mat. We will save that story for another day. The story of the invoice typically goes something like this: “Nah, I don’t need it.” Or you grab it with good intentions to file it away and before the end of the day you’ve used it to wipe grease off your hands. It then goes to the floorboard to die a slow and painful death.

But that’s valuable information

A few weeks ago I kept thinking how nice it would be to take a picture of that invoice before the grease hits it, send that picture as a text message, and get it applied to your machinery automatically. So I built that system.

I used Twilio to create a phone number that can send and receive messages. It costs me about $5 a month but I now have a direct line where I can text information and send pictures that are valuable to my operation. If I want to “log” an irrigation event, all I have to do is text which field I watered. If I plant a field, I can text what field was planted and which variety. When the bean header breaks down, I can take a picture of the invoice and send it to my software.

The picture below is what lands in the software.

As you can see, the software incorrectly guessed which piece of equipment this invoice was tied to. Whose fault is that? Actually mine. The reason is that I didn’t have a specific name for my bean header in my software at the time I sent the invoice. I had a generic name— 30 foot bean header— and I wrote 2152 Case IH bean header on the invoice. I ended up changing the name of the bean header in the software so the problem doesn’t happen again.

At the end of the season, I’ll know exactly what I spent on the header in repairs and maintenance. In December, I’ll be able to run the numbers to see if I should upgrade or keep what I’ve got.

Find your weakness

Part of my goal in using AI is to find ways to improve my weaknesses. I am not great at record keeping. I can’t tell you how many invoices have died in the floor of my truck and how many applications didn’t get recorded. But one of the reasons I am building my own software is to help automate the tasks where I am genuinely weak.

I can send a text message. And if that’s the difference between good data and bad data—or no data at all— it’s worth my time.

If you are interested in building something of your own, reach out. I’d love to talk about it with you. As for me, I’ll be finding more weaknesses to address.

Leave a comment

Subscribe now

Share

If you’ve ever talked to a farmer during planting season, you know how important it is to get the seed in the moisture. Plant the seed too shallow, and it sits in the dry dirt until the next rain arrives.

Get it in the moisture, and the magic happens. First the seed absorbs the water surrounding it. Then it sends a root down to help push. Eventually, a new thing emerges.

The first month is slow because most of the action is happening under the surface. That’s where the roots are trying to make trades. Eventually the roots will leak a chemical into the soil to let the bacteria know it’s ready to make a deal.

When the bacteria responds, the plant opens a tunnel through its own root tissue and lets the bacteria walk right in. Hospitable little roots they are.

The soybean plant feeds the bacteria the sugar it creates through photosynthesis. The bacteria provides Nitrogen for the plant to grow.

Around the end of May, my soybeans were flowering.

Soybean flowers are small—about the size of pencil eraser. Some of those flowers pollinated and some of them did not. I still have a lot to learn about why some of the flowers turn into pods and some of them don’t.

During the growing season, the soybean plant knows that it will need phosphorus to fill the seeds in the pods. So it puts out a signal to the fungi in the root zone. The fungi gets fed the same way that the bacteria does.

By the time harvest rolls around, which is now, the plant has shed all of its leaves and given all of its energy to the seeds.

What comes out of the combine this week will look almost identical to what I put in the ground in late March. I guess the one I sunk 2 inches into the soil in March was waiting for life to begin. The one coming out is carrying the work of roots, bacteria, fungi, sunlight, and rain— though maybe not as much rain as I would have liked.

Some of those seeds will go on a barge down to New Orleans. Some of them will get crushed for oil and meal. And some of those seeds, somewhere, we be planted again.

Five months ago I planted what looked like a lifeless seed. Today, I begin harvesting a season.

In honor of harvest beginning on my farm, I am sharing my green bean bundle recipe. No, we don’t eat the soybeans we grow. But we do eat the green beans from our garden.

Subscribe now

Share

Harrison was afraid that touching the switch might kill him.

The fear started about five years before lights were installed at the White House. At the time, there were two companies racing to bring electricity into American homes. Westinghouse Electric was building a distribution network on alternating current and Edison Electric was building one on direct current.

During the race to electrify America, Thomas Edison wrote to one of his executives: “Just as certain as death Westinghouse will kill a customer within six months after he puts in a system of any size.” He wasn’t making a prediction. He was starting a campaign.

In February 1888 Edison’s company published an 84 page pamphlet titled “A Warning from the Edison Electric Light Company,” collecting every alternating death story it could find. That summer an Edison ally electrocuted dogs (and other animals) in front of the press at Columbia College. They’re killing the dogs! Edison partisans pushed “Westinghoused” as the verb for dying in the electric chair. The whole documented sequence is here.

Then, on October 11, 1889, a Western Union lineman named John Feeks touched a high voltage line in Manhattan and burned in front of a large crowd. New York decided to put their wires underground after that incident. Edison ended up getting the White House job.

So put yourself in 1891. The most famous inventor in America is telling you this new technology may kill you and it can’t be trusted. Are you going to flip the switch?

It happens a lot

Every time a new technology is developed, a version of this process plays out in real time.

A Nature editorial in 1889: “The telephone is the most dangerous of all because it enters into every dwelling. Its interminable network of wires is a perpetual menace to life and property.”

In 1906, before he was President, Woodrow Wilson accused the automobile of spreading socialism because it pitted the working class against the rich. The carriage makers and horse owners created ads to sway the public view about how unsafe and unreliable the automobile was.

Fast forward to February, 1962, and a reporter asked President Kennedy about Labor Department estimates that machines were displacing 1.8 million workers a year. Kennedy answered: “I regard it as the major domestic challenge, really, of the ‘60s, to maintain full employment at a time when automation, of course, is replacing men.”

The next article could run tomorrow in the New York Times, “One of the greatest problems with automation is not the worker who is fired, but the worker who is not hired.” It’s a Time article from 1961.

In March 1964, thirty five people including the Nobel laureate Linus Pauling sent President Johnson an open memorandum arguing that “the cybernation revolution” had permanently broken the link between jobs and income, and that the government should therefore guarantee every family an income as a matter of right. Sound familiar?

The internet needs no introduction. In September 2000, Pew asked Americans who were not online—why. Fifty four percent said the internet is a dangerous thing. About 31 million American adults said they would never go online. Not later. Never.

It seems that every technological revolution has to have its fair share of hype and fear at the beginning. This time is no different.

Who’s holding the pamphlet now?

Edison wasn’t lying when he said that high voltage alternating current will kill a human—his allies even demonstrated it would kill animals. But what he did was take a real fear and amplify it so that he could take advantage of it. He owned the direct current patents. The 84 page pamphlet was a business decision.

Somebody is always holding the pamphlet.

In May 2025, Dario Amodei, the CEO of Anthropic, told Axios that AI could wipe out half of all entry level white collar jobs and push unemployment to 10 or 20 percent within one to five years. Sam Altman said this July that “we are now, like, in the singularity.”

Singularity? What he means is that we are in the period of time where technology changes society so fast that even our short-term predictions about the future are now useless. This doesn’t feel like a singularity to me.

So why would two leaders of the fastest growing businesses in the history of the world say all of those things? Fear sells. Hype sells.

OpenAI raised $122,000,000,000 in March. Yes, $122 Billion. Anthropic raised $65 billion in May. How do you raise that much money? You spread fear and hype.

Edison had reasons to want people afraid of alternating current. These men want you to fear being left behind.

The invention comes first. The payoff comes later.

And to be fair there is always displacement when a new technology arrives. The telephone put pressure on the telegraph business. There’s now about 300 million automobiles in the United States and just a few carriages left in New Orleans. Automation eliminates jobs. The internet shifted information sharing to a digital platform and moved jobs along the way.

But with each technological revolution, new jobs are created in its wake. Even with automation we see that people move to more productive parts of the economy with a focus on building, managing, and improving technology.

A federal commission studying automation reached a better conclusion in 1966: “Technology eliminates jobs, not work.” Jobs will shift, and new jobs that we could have never imagined will be created. Influencer, anyone?

But historically, the change takes time.

In 1990, Stanford economist Paul David explained the situation in a paper called “The Dynamo and the Computer.” Electricity had been available for decades before it produced a meaningful leap in factory productivity. Why? The factory.

The factories of the time had been designed to run on a steam engine turning one long driveshaft. Belts ran from that shaft to the machines on the assembly line. Replacing the steam engine with an electric motor left everything else unchanged. The power source was new, but the factory was still organized for the world powered by steam.

The payoff came when manufacturers stopped asking electricity to run the old factory and began designing a new factory around what electricity could do. They put a small motor on each machine. They changed the buildings, rearranged the production line, and reorganized the work.

The invention was important. The reorganization created the value.

That is where we are with AI

A farm can buy an AI tool tomorrow. But if the field records are scattered across notebooks, text messages, files, and one person’s memory—the tool has little to work with. The AI model may be new, but it won’t matter if your operation is still organized for the old world.

Maybe that helps explain why 52 percent of farmers surveyed by Purdue said AI and data-driven tools currently provide no meaningful benefit to their operations. I don’t think that means that AI is useless. I think it’s evidence that access to a technology and the ability to benefit from it are not the same thing.

In the world of immediate gratification, most of us just want to buy the tool that solves our problem. But you can’t “buy” your way into the future. It doesn’t matter how awesome the tools get if the tool doesn’t have access to your data and your context. The payoff comes by digitally recreating your context and layering the technology on top.

We know how this feels and we experience it all the time. Pretty much every new tractor is now shipped with auto steer. Before long, the see and spray sprayers will be the used 6 year old machine you buy at an auction. It takes time for technology to disperse throughout the economy.

Be curious, not judgmental

President Harrison didn’t immediately trust the new technology and I think it’s normal to feel that way. But eventually, a generation grew up flipping the switch without thinking about it.

AI will likely follow a similar path.

But don’t believe everything you hear coming out of Silicon Valley. The world’s not going to end if you don’t reorganize your operation around a chatbot tomorrow. And you certainly do not need to buy something because somebody has frightened you about being left behind.

I think now is a great time to ask questions.

This week, find a task that you hate doing; grain tickets during harvest, an upcoming lease renewal, keeping up with the markets. Write down what that task costs in hours, dollars, delays, and mistakes. Then ask what would have to change for that work to become easier.

Do not start with AI. Start with the work.

I think that’s the lesson we can learn from the tech revolutions in our past. The winners were not the people who predicted the future most accurately or bought the first dynamo. They were the people who understood their operations well enough to recognize what the new technology made possible—and then rebuilt around it. Brick by brick.

Subscribe now

Share

Lately I have been getting up before the sun. Ever since I had little ones I have trained my body to wake up early to try to find some peace before the chaos. By the time I get my coffee and get cozy on our screened in porch, the sky is just turning pink. It’s still too dark for me to see my books, and the sky demands an audience, so I just sit quietly and give myself a moment to be in awe and to not think, and just to behold..

Then I watch as everything wakes up.

Remi, our Great Pyrenees, is the only one already up. She’s been awake all night barking intimidatingly at anything that dares move a muscle on our property. Remi’s job on the farm is to guard the chickens who, without her, have a bad habit of getting slaughtered by the foxes, coyotes, raccoons, and occasionally the neighbor’s dogs.

Two chickens are rounding the corner as the sky is lightening, staying close to the house, heads low, pecking around for treats. Those defenseless birds are probably my favorite animals to watch. They have a very easy life now with their guardian in place. The door to their coop stays open most of the time, so they spend their days wandering around our lake front property pecking at grass, garden scraps, bugs, and the leftovers from our fridge (job #1: food waste recycler). They are social creatures, so they stick together in groups of 2, 3, or 4 and they talk to each other (and me sometimes). The rooster, Rex, is always making sure everyone is okay. Whenever he finds food, he clucks at his hens to get their attention, picks it up and sets it back down so they can see it, and only ever eats after the ladies have begun. (This is a real rooster thing, and it’s called “tidbitting”.)

The mama hens are the same with their chicks. Around springtime one or two of them will go “broody,” her body instincts telling her it is time to incubate her fertilized eggs. She will pick a comfy box in the coop, pluck out her breast feathers so she can keep the eggs close to her skin, and sit on them for 21 days, only taking short breaks to get food and water. When they hatch, she guards them with her life, charging at any human or animal who gets close. She sits on them the first few days, but it isn’t long before they are out on the yard with her and she is clucking to them to show them how to find food or keep them close by and safe.

And of course, the chickens play their role in providing us with our daily protein packaged in a naturally shelf stable container coming in the most beautiful assortment of browns, creams, blues, and greens (job #2). If you take a minute to sit and think about that, it will darn near blow your mind.

When you spend time in nature, you begin to realize how naturally all the pieces seem to fit together. Everyone has a job. The trees use their leaves and photosynthesis to pull energy into the plant. When the leaves die and fall to the ground bacteria and fungi break them down to create organic matter to feed the soil. The healthy soil takes in the seed and the seed grows into a plant for food. The tree and the soil become habitat for all kinds of insects and animals that create an intricate food web. I could go on and on…

Living on the farm, we have space to observe these relationships. About seven years ago, we were introduced to regenerative agriculture. In my interpretation, regenerative agriculture is an attempt to observe those natural systems and relationships and farming in harmony with them. It isn’t easy, and we have a long way to go, but it feels good to watch, to learn, to appreciate and to have a vision to work towards.

I find so much joy in observing all the seemingly disparate parts of nature playing their role in this intricately designed system. Even the chicken poop plays an important role, but we will save the subject of poop for another day.

For now, I will leave you with a recipe that allows us to use all of those beautiful eggs sitting on our counter. I hope you try it. And be sure to get your eggs from a friend or a farmer with chickens. And if you can, go sit a while and watch those miraculous flightless birds playing their role so well.

Subscribe now

Share

But what most of those tourists don’t know is that Clarksdale is also where the machine to pick cotton was perfected.

I grew up passing Hopson Plantation every day on my way to school. Hopson is located just south of Clarksdale on highway 49, and in the 1930’s, International Harvester used it as a test farm for the spindle picker. On October 2 of 1944, eight machines picked a field in a single day. It changed the world.

But guess what, the Hopson brothers who ran the plantation still had to make the math work. It’s great that the machine could pick more acres than a human and save time, but what was the cost? So they ran the numbers. According to their own math, hand picking cost $39 a bale. The machine cost a little over $5.

I wish most of my decisions were that easy. Dollars, and not cents. But here we are.

Turn row math

Every year, we farmers with storage capacity face the same decision at harvest. Sell it out of the field, or put it up and wait. What’s the true cost to store our grain?

Let’s say you are cutting corn right now. On Monday, August 17, my local elevator is bidding $4.71 for August delivery corn. In this scenario, let’s pretend you have room for 30,000 bushels, and the same elevator will pay $5.2175 for January delivery.

It sounds like a no brainer, right? But let’s estimate that storage costs you an average of 3.5 cents a bushel a month, and by January you will have accumulated 17.5 cents of storage costs per bushel.

So far, we’ve estimated that you will need $4.885 ($4.71+$.175) in January to break even. But there are a few more data points that we need to add before we make a decision.

The first is interest. Sell today and $141,300 goes in the bank this week. Store it and that money sits in the bin instead. If your operating note costs 7.5 percent, five months of carrying that money is about 15 cents a bushel. A little over $4400.

The second is shrink. Corn loses weight in the bin mostly in the form of water. Half a percent sounds like nothing but it can add up on bigger numbers. On 30,000 bushels it is 150 bushels that won’t exist because of the extra drying that occurs. The math is a bit more complicated that that, but for the purpose of this example we will keep it simple.

The last number that we will consider is extra hauling costs. If you are having to ship the corn twice, there’s an added cost. I used .06 for the purposes of this illustration.

When you add it all up the real breakeven is about $5.12, not $4.885.

So what should the farmer do in this situation?

He should store it if he’s going to price today.

The January bid only clears breakeven by about 10 cents, but that is roughly $3000 on 30,000 bushels. A 51 cent spread that felt like free money shrank to 10 cents once you counted everything. And guess what, by the time you are reading this the prices have already changed and so has the math.

So I built a calculator. Who wants to run all those numbers again today?

Here it is: Sell it or Store it

Type in your bushels, today’s bid, months you would hold, storage cost, interest rate, shrink, and any hauling difference. It tells you the price you need in the future, illustrates where every cent of carry goes, and lets you slide the months around and watch the interest stack up.

I built that tool with Lovable— and in about twenty minutes and 3 small prompts—I had a working application. I described what I wanted in plain English, it built it, and now I have a tool that I can use throughout harvest.

Lovable works in the browser—there’s nothing to install. The free plan gives you a small number of credits each day, and every message you send burns some of your credits. Sell it or store it didn’t cost me a dime because it was such a simple build.

And when you are done, you hit publish in the top right corner and it hands you a link on their domain that you can save to your phone or send to a friend.

This link gets you 10 free credits on Lovable

The times they are a-changin’

In 2023, building software required deep fluency in programming syntax, required you to manually write a bunch of boilerplate code, and hours of debugging. In 2026, you can build your own software in an afternoon. Using plain English.

Last week I showed you the system I am building, the software that catches my rainfall and tracks my irrigation the way a yoyo catches fish while I am off doing something else.

That’s the way I like to build. I think about the things that I hate doing, and I try to automate those tasks. A tool is only worth having if it saves time or money.

Go try it

There are probably a thousand small tools that you can build to help with decision making. Maybe it’s a budget application tailored to your life, business, or farm. Maybe it’s an application that stores pictures of your invoices and files them in a folder for you. Maybe it’s an equipment maintenance and hours tracker.

But before you trust any tool on a real decision, make sure you’ve checked behind it to make sure the math is correct. I checked the tool I built on Lovable with a pencil and a piece of paper. If it was something that I was building for my own software, I would’ve spent hours researching the best way to build the tool. I built this one just to show you what is possible.

One of the greatest achievements of the AI era is that it has democratized software creation. You can build the tool for your operation. For your business. For your life. Maybe that side hustle gets a website.

It would absolutely make my day to see some links in the comments of what you guys are building. Happy building!

Sources and further reading:

Lovable 10 Free Credits

Hopson Plantation

Subscribe now

Share

Leave a comment

I was just a little boy on my first trip to the river. What I remember is a tent on a sandbar, a driftwood fire, and huge fishing poles. I also remember leaving the poles out all night, equipped with small bells to alarm us when the poles started moving. It was like Christmas morning when the bells went off, you never knew what you were going to get.

These days, I get to take my girls to the river. You’d be surprised how much I learn about them there. “Are there alligators in here?” Oh yeah! “Are there snakes, daddy?” Yep! “Are there sharks?” I doubt it, but they did catch a bull shark around St. Louis some time ago. Or at least that’s what I was told.

And they always get in.

Thirty years later and the current still pulls my eyes across the water. Occasionally, I’ll see an uprooted tree calmly passing by—almost as if it belonged there. Sometimes I let my mind drift down to the river bottom, to this very sand—thirty feet underwater come spring.

My youngest is seven. The other day she waded out a step too far and an eddy grabbed her. I let her swim until I saw the panic in her eyes. It wasn’t easy fighting the current and the rocks with a small child on my back, but we eventually made it back to the bank. She sat a long while and stared out across the water. Sometimes I stare too.

I still bring a pole sometimes. I watch the girls play. I watch the river go by. I rarely bring a book, but it's the place I read the most.

When the bell rings

Here’s a recipe for the river monsters that ring your bells. Yes, the first two recipes of this series are both fried. This is Mississippi. We didn’t weigh the fish pictured above—but that same day we caught a 42 pound flathead.

We vacuum sealed what we didn’t eat fresh. If you don’t have a vacuum sealer, you can fill a ziplock bag with water and drop in your filets. Food doesn’t get more fresh than this—and you should see the smiles on those little faces when they eat something that they caught. You can’t beat it with a stick.

Share

Our yo-yoing adventure normally begins with a trip out into the fields to collect some earthworms. Then we launch the boat, bait the lines, and head off to swim for a while.

If you have never used a yoyo, it is a spring-loaded reel that sets the hook for you and holds tension so the fish stays on the line. You can put out a lot of hooks at once, which makes it easy to catch a mess of fish in a short period of time. It’s a great strategy if you are planning to cook for a lot of people.

But even though the yoyo does the hard part, it still needs a human in the loop. You still have to run the lines, take the fish off, and reset the bait.

A yo-yo is actually a small system. The spring, the nylon line, and hook all work together to accomplish the goal of catching a fish.

That is what we have been building for the past nine weeks. Not a collection of AI magic, but the parts necessary to build a system that can keep working for us after we step away—while still leaving the judgment to us.

In the first issue, I talked about providing context so that we could get better answers from the large language models.

Issue 1

In the second issue, we customized our instructions so that the LLMs had context about us that applied to every discussion.

Issue 2

In the third issue, we built our first project so that we could provide context for our farm or business.

Issue 3

The fourth issue was an introduction to my life and the agricultural treadmill that a lot of farmers find themselves on.

Issue 4

In the fifth issue, we gave our project instructions like we would an actual employee.

Issue 5

The sixth issue was where we made the “employee” work autonomously. We set up an autonomous workflow where the AI fetched the weather, the markets, policy, and even checked our email for us.

Issue 6

In the seventh issue, we talked about capture. To get good results with AI, we have to capture the right information and store it in the right place.

Issue 7

In the eighth issue, we started pointing AI towards our data in cowork—because the AI has amnesia.

Issue 8

In the ninth issue, we discussed how we can use the research function to find data for us.

Issue 9

I think every issue can stand alone and be useful for your farm or business. But today, we are going to explore how we can connect it all into a system.

We are building a system.

All of the issues are part of a system that we’ve been building for weeks. I now want to show you how I decided to set up my system.

Instead of building a thousand small tools to help with decision making, I decided that I wanted to build one system to bring all the tools together. The reason is simple. I want all of my data in one place that I control.

The system that I chose was software. I chose software because it allows me to connect all of my “silos” in one place and automate a lot of tasks that would otherwise require lots of time.

Below, I am going to illustrate one part of the software that I’ve built—this one feature of the software captures 640 data points per day. That would be incredibly painful to do manually.

The system, assembled

First of all, software is not as complicated as it sounds. There are a myriad of tools now available that allow you to build software very cheaply. If I can do it, you can do it too. For the most part, I’ve used free resources to help build the software. Here is a screenshot inside of the software where I’ve mapped out my field boundaries and labeled them based on the irrigation. The green is row water, the yellow is pivot irrigated, and the red is dryland.

I also have budgets, field data, government program data, soil maps, etc. But, for the purpose of illustration, let’s take a look at just one feature today.

One of the things that’s hardest to keep up with as a farmer is rainfall. There are some decent apps out there that will keep track of your estimated rainfall, but that’s about all it does. It estimates how much rain fell at one location.

That can be useful, especially if your farm is spread out over a bunch of miles, but the data is only valuable for a day or two. It can answer questions like “which field is dry enough to spray?”

But I wanted a rainfall application where the data was valuable indefinitely. So I built it.

The current version of my application estimates rainfall amounts on all 49 of my fields. It logs that data. Then, I verify that data. I have actual rain gauges scattered throughout the farm. If the digital rain gauge is showing I got an inch, but the actual amount is half an inch, I can go in on the software and override the estimated value. If you see a little green clock next to the amount, that means that I went in and changed the estimated value to the observed value. This is a picture of what one field looked like during the month of July.

And to be honest, I hate capturing data. But the system I built can run without me. It’s the yo-yo that catches the fish. Among other things, the software logs estimated rainfall data and high and low temperatures for every field—every day. I am only there to correct after the data has been logged.

I also built a cumulative rainfall tracker to help make irrigation decisions. Because I enter my planting dates into the software, the tool can track what growth stage the plant is in and estimate how much water the crop is using at that stage. The software notifies me when I have a rainfall deficit and I can then decide whether or not to act on that recommendation.

How I actually make the call

Every one of my irrigated fields has a card like this. Now that I am capturing rainfall, evapotranspiration based on each individual crop and its stage—plus my actual irrigation events—I have a full picture of what is happening on the field level.

Instead of the data being there for a week, the data persists throughout the growing season and it becomes another layer for decision making.

What can you build?

You can build your first software tool today. Replit and Lovable make it super easy to get started. You don’t need to know how to code. You don’t need to worry about hosting, frontend, backend, or database. They handle all of that. You describe what you want built, and it handles the rest. But don’t forget to include your context and clear instructions, you’ll get much better results.

A few months ago I had an app on my phone that would track my estimated rainfall. It was called raindrop and it did a decent enough job. But that rainfall data wasn’t connected to anything else. It didn’t know what variety I planted or how much I irrigated. It didn’t matter if I logged my actual rainfall amounts because it was just a silo of data, something to look at when I was bored.

Now, that rainfall data is helping me make better decisions in season. It’s a stored log connected to my irrigation data. At the end of the season I can now see which dryland fields got more rainfall and how much that contributed to yield. The more data I have on rainfall and irrigation, the better equipped I will be for capital expenditure questions that arise in the future. Does it make sense to add irrigation to this field?

But the goal is not to build software that makes every decision for me. It is to build a system that is working for me while I am busy farming.

The tool can gather the rainfall, track the heat, remember the irrigation events, and show me where the deficit is. Then I’ll do the part that matters most: make the call.

It becomes a loop. The data persists. The lessons start to compound with each passing day and each season. It’s no longer just a memory. It’s stored information that we used to make a decision.

The yo-yos catch fish because they were designed to be a system that works while we are away. And we can leverage AI and software to do the same. We can let it work while we sleep, then we’ll show up when it’s time to make a decision.

Subscribe now

Share

Two barn swallows are floating through the morning air, feeding the next generation of barn swallows that have made my porch their home. The chickens are out early too. Wobbling around the yard like mini T-rexes. Millie the dog, has made it her intention to ruin this moment for me—but I won’t allow it.

There aren’t many mornings this time of year where I choose to be outside just to be in nature. There’s too much to do. Too much to learn. At least that’s what I tell myself. But as I’m writing this, I’m learning why I love the winter so much and perhaps the reason I’ve been in a funk lately.

The winter is the time of year when I get to turn my brain off for a season and when I get to be in nature the most. Long walks in the woods. Hours of contemplation on a deer stand. Camaraderie in the cypress break.

One of my favorite animals to watch in nature is the wood duck. I get to watch them from the deer stand sometimes. I love how playful they are and how their heads bob like they are living to a beat.

Perhaps nature allows me to think about myself less and removes me from the center of the universe. It’s odd how freeing that thought is for me.

The egret hasn’t moved much, just a white spot across the lake pressed against green cutgrass that will soon be—brown.

It’s deep summer here—hot and humid—the time of year when the weeds in our garden have pretty much taken over. But there’s always the last plant standing in our garden. Still producing when all the others have stopped. It’s a close race this year against the peppers, but I suspect the okra will win again.

I can still remember the way my grandmother would layer the paper towels in the bowl before she pulled them from the fry daddy. The contrast between the white, green, and brown still seared into my mind.

I still love to stand over the bowl and take in the smell while the okra cools. And yes, the crunch still hits the same 30 years later.

This is fried okra. Gran’s way. I hope you enjoy.

Subscribe now

Share

The flyer looks more like a tent than a modern day airplane. Cotton cloth, wood, bicycle chains. The Wright Flyer isn’t impressive to the naked eye with our modern standards of flight, but the story behind the Wright Flyer is truly incredible.

As you probably know, the brothers succeeded in December of 1903 with the first human flight. But in August of 1901, Wilbur Wright allegedly told his brother that man would not fly for fifty years.

The reason was that the Wrights had just spent a second season at Kitty Hawk, North Carolina with a glider that didn’t work. Their 1901 machine was generating roughly one third of the lift their calculations said it should. They had also spent the last few years of their life building, hauling, and crashing their newly built creations. Frustrating.

I think most of us would have given up. They didn’t. They dug in.

The problem was the brothers had designed their wings using the standard lift equation of the time, the one every aeronautical experimenter in the world was using. Inside that equation sat a value called Smeaton’s coefficient. A number describing air pressure that had been accepted since the 1700s. Everyone used it. Otto Lilienthal, the German glider pioneer who published the wing data the Wrights had built on, had used it. It was not controversial. It was accepted.

It was also wrong.

Lilienthal’s own measurements were sound. Later analysis suggests his data was accurate within the limits of his own instruments. It simply couldn’t be transferred to a wing configuration different from the one he had tested. The failure of the Wright brothers Flyer was a shared constant that nobody in the field had recalculated in a century and a half, sitting quietly inside everyone’s arithmetic, scaling every answer down by a third.

But after a few failures, the two bicycle mechanics from Dayton, Ohio decided to recheck the math.

What checking cost, and what it bought

From September to December 1901, the Wrights built a wind tunnel. A wooden box about six feet long with a fan pushing air through it.

They ran as many as two hundred wing shapes through it, one at a time, and recorded the data on nearly fifty of them. They measured lift. They measured drag. And by December they had done something no one else had. They had generated their own aerodynamic data, from their own instruments, for their own wings.

They also found a new number. Against the accepted 0.005, they derived their own Smeaton coefficient of roughly 0.0033.

A few years later, in December 1903, Orville Wright piloted the first powered flight with a 12 horsepower engine. At the exact same time, other people around the world with much larger budgets, continued to fail. So why did they succeed?

The Wright brothers super power was research. Research didn’t help them fly faster. Research is the only reason they flew at all.

Research, by definition, is careful systematic investigation and study. The word is derived from the Old French term recerchier, which means to find new facts and test ideas.

What I’ve discovered over the past few months, is that AI is pretty good at research. I’ve used the research function in Claude to help find some of my opening stories in this substack series. I’ve used it to determine whether or not I should apply a fungicide on my soybeans. I’ve used it to build specs that I can plug into Claude code for the software that I am building for my farm.

The research function of these large language models is one my favorite features. Today, I’m going to show you how I use the research function, and why I think it deserves a place in your toolbox.

Let’s begin

To illustrate the research function, I picked three different soybean varieties that I actually planted this spring. Asgrow 49XF4, Revere 49F36, and Revere 47F77. I asked the exact same question in the general chat function of Claude and again with the research function turned on in my Long Lake Farms project.

This was my exact prompt: “I planted Asgrow 49XF4, Revere 49F36, and Revere 47F77 soybeans this year. How should I think about managing these soybeans during the growing season this year?”

So what did the basic chat say? So glad you asked. Some of the results were generally useful. It told me that the highest ROI decision right now is around irrigation timing. It told me that red banded stink bugs can rob yield and that I need to be checking for them in my fields. It also told me, at about 80 percent confidence, that the Revere soybeans had the Enlist trait. They aren’t Enlist soybeans and the chat was wrong.

You can read the full response here

So why did the AI give a wrong answer?

I’ll spare you all of the details about the models reasoning process and where it went wrong. But the premise was that it made the wrong connection and then built it’s entire response on top of that wrong connection.

The model thought Revere soybeans were actually an old Terral distributed seed line called REV, which was retired a few years ago. Revere Seed is a completely different company. The model matched two similar sounding brand names, inherited the wrong brand’s herbicide platform, and then built most of the response on top of the wrong product. Split chemistry. Cleanout protocol. Adjacency risk. Once again, you can see the entire response up above.

And the answer to what I actually planted was sitting in a PDF on Revere’s own website the entire time. A table with my variety names in the left column and the word XtendFlex three columns over. If I didn’t know any better, and didn’t check behind the AI, I could’ve killed an entire soybean crop by spraying the wrong chemical.

That is Smeaton’s coefficient. A plausible inherited value, never checked, quietly scaling everything downstream.

What happened when I ran it as research instead

I ran the exact same question through my Long Lake Farms project, and I hit the “+” button to turn on research before I asked that exact same question: “I planted Asgrow 49XF4, Revere 49F36, and Revere 47F77 soybeans this year. How should I think about managing these soybeans during the growing season this year?” Turning on research tells the model to go check instead of guess. Running it inside a project gives the model additional context that will lead to better answers.

Here is an article where I describe context

First of all, it asked questions before it answered

Before it answered the question, it asked me two clarifying questions that helped with the research.

The first question it asked was about timeframe, and I told it to do full season research. The second question it asked was whether I wanted the answer anchored to my actual fields.

The normal chat that almost everyone uses never asked me anything. I gave it three variety names and it handed me a plan that was only partially useful. The research function in my project paused and asked whether I wanted a general answer about the three soybean varieties or a specific answer about the ground I am standing on. Obviously, if I was actually running the research for my farm, I would have told it to build the document for my actual fields. But for the sake of comparison, I told it to not anchor the response to my fields.

It could only ask that because it was running inside the project. The project gave it the context it needed. A cold chat has no idea your fields exist, so it cannot offer to use them. It guesses and moves on.

If you are new to my substack, I talk about Projects here

That is the difference between a tool that knows farming and a tool that knows your farm.

What did the research inside of my project deliver?

It got the trait platform right, which is very important and the exact thing the fast chat got wrong. Then it found things the fast answer never knew to look for. The Asgrow bean is a chloride includer. Both Revere beans are excluders. That is a placement decision that could actually make me money.

On fungicide, the fast chat guessed that a fungicide application delivered roughly two to four bushels. The research run pulled a meta analysis of 240 trials and came back with 2.7 percent, about 1.64 bushels per acre, and named the conditions that make a fungicide application worth it. One of those numbers is a guess. The other is 240 trials.

And it gave cited irrigation advice from Mississippi State, with a real trigger. Terminate irrigation at R6.5. It even had the MSU number showing that “trigger” lifted water use efficiency 36 percent with no yield penalty. That’s a big deal.

You can read the researched response here

What this means for our work

You do not need the research function for every question. Most days you want the fast answer, and the fast answer is fine for most questions. Research is much slower and it burns a lot more tokens, so save it for the questions where being wrong is expensive.

The research function also requires a paid plan, and you can hit your usage limits pretty quickly. I like to use the research function when I know I am going to be away from the computer or phone for a while, so that when I come back, I have a document that I can read through to begin the next task.

When the research is done, I can summarize that information and add it into my project context. I’m gaining more context about my farm and the products that I am using, and so is the AI. It’s a feedback loop that is pretty powerful.

The purpose of research is to give us information for decisions and actions that haven’t yet occurred. You give your time and energy now to lower the cost of being wrong later. And sometimes, research overturns a belief you never checked. I use research to assemble scattered facts into something that I can act on.

The Wright brothers didn’t fly because they began with all the answers. They flew because they did their own research, and used that new information to change the way they were working. They built a box with a fan and rechecked the math that everyone assumed was correct.

Research is our opportunity to stack the facts before we bet the farm.

I’m genuinely excited about this new era, where gathering information has become more accessible and less time consuming than ever before. And the real fun begins when we see that research pay off on real decisions.

I hope we will all be flying soon.

Sources and further reading:

More information about the Wright Brothers

More information about Claude and Research

Ethan Mollick, "The End of Search, The Beginning of Research"

Share

Subscribe now

Fast forward to my senior year of college, and I was still incredibly disciplined. The discipline had just found new hobbies. I retired from track. I played basketball every day at the Turner Center. I went to the bar at least three times a week and became a world class shuffleboard player. And I only attended the classes that required attendance. One of those classes that didn’t require attendance was Finance 361: Financial Analysis and Modeling in Excel. I was there the first day of class, I was there for the midterm, and I was there for the final. That was it. I can still remember the all night cram sessions before the midterm and the final, watching recorded lectures from six weeks earlier. I made my first C in that class, but C’s do in fact get degrees. I will say that Finance 361 is the only class in college I regret skipping.

After college, thanks to my wonderful time in the Finance 361 classroom, I became a spreadsheet junkie. I now have over 300 spreadsheets covering everything from crappie fishing to farm profitability. For the past few weeks, we’ve been talking about data and the filing systems we can use to organize it. For the last 16 growing seasons, I’ve used spreadsheets to organize my data.

One of my favorite spreadsheets I’ve ever built, and one I still use today, is a field by field prospective plantings engine that tells me which crop is most profitable to plant in each field. It takes my historical yield data for that field (or expected yield), current prices, my expected variable input costs, and it tells me what crop delivers the best return over variable cost. Not what the coffee shop thinks. Not what my gut says after a big cotton crop. What the numbers say, field by field. The Long Lake field might pencil out to beans while the home place begs for corn, and the spreadsheet doesn’t care which one I feel like planting. I’ll provide a link down at the bottom if you want to try it out.

But naturally, I tried to replace it.

A few weeks ago I sat down with Claude Cowork and described the entire spreadsheet. I’ll explain Cowork down below. I told it what it did, and I dropped the spreadsheet into the chat. In about ten minutes, it built me a working tool. The tool worked great and it was definitely prettier than anything I’ve ever built in Excel. I felt like I’d skipped Finance 361 all over again but this time aced the final.

Then I closed the laptop.

When I came back the next morning, the tool was still there. The numbers I’d punched in were gone. It had done the math brilliantly and remembered none of it. It was a genius with amnesia. Every session started from the hardcoded values in the artifact and not my actual farm information.

My ugly spreadsheet, the one I’ve used for 15 years, remembered everything.

We’ve spent the last month talking about data. Backing it up, organizing it, keeping it where you can find it.

A tool is only worth what it remembers. Amirite? I’ve been beating this dead horse for a reason…

The flashiest AI on earth loses to a fifteen year old spreadsheet if the spreadsheet holds your history and the AI doesn’t. The AI doesn’t know the yields from 2022, how much it rained, or the notes in your journal. That’s the asset. The AI is just the hired “mind” you point at it.

But before you feel too good about that, I have to tell you what my spreadsheets can’t do, because this is the part that’s been eating at me for years, long before AI showed up.

They can’t talk to each other.

My prospective plantings engine knows expected yields, prices before planting, and variable costs before the season begins. Meanwhile, three folders over, there’s a spreadsheet with actual plantings, varieties, and yield observed during that year. The two of them have never connected. Not once. The actual plantings spreadsheet knows what was planted on the home place in 2022. The prospective planting engine only told me which field delivered the most over variable cost before I planted in 2022. Anyone ever changed their plan during planting season? I laughed while I typed that.

The same is true with the satellite imagery I save every week during the season. Those NDVI maps can see a sick spot in a field two weeks before I notice it from the truck. But that file can’t whisper a word to my expected yields sheet. The yield sheet finds out the crop was struggling the same way I do most years. At the scale, or at the gin, in October, when it’s too late to do anything but bitch and moan. Boy, we farmers are good at that.

The point is that AI is a brain with no memory. Your data stores the memory but has no brain. Connecting the two is where the magic happens.

There’s a show on Apple TV called Silo that Cali and I have been watching. It’s based on the Wool, Shift, and Dust Trilogy by Hugh Howey. The premise of the story is that there was an event in the past that sent thousands of people to live underground in Silos because the earth’s environment has become uninhabitable for some reason. We find out during the story that there are around 50 other silos but nobody knows the others exist. The AI algorithm that functions as the leader has set up safeguards (it will kill them) to make sure that each silo can’t communicate with the other silos. Everybody stays sealed in, certain their silo is the whole world.

That’s kind of what a filing cabinet is. Fifty silos, each holding a piece of the truth, forbidden from comparing notes.

Data in a silo isn’t really actionable, because it’s only seeing part of the picture. My prospective plantings file can’t communicate with my actual planting data. My NDVI satellite images that I gather throughout the season are only valuable when I look at them. The rainfall data on my Raindrop app doesn’t connect to my actual fields. The tissue samples that I pull have no idea what’s actually in the soil from the place I pulled the sample. My yield history can’t begin to explain itself. The benefit that we are missing as farmers lives between the files, in the connections none of the spreadsheets can make. The tunnels. I don’t have a data problem. I’ve been collecting data like a madman since Finance 361. I have a tunneling problem.

The problem wasn’t cowork. The problem was the job I gave.

I asked it to become the memory. To hold my numbers inside a tool it built from scratch. That was always going to fail, because the memory already existed. It was sitting in the folder on my laptop where fifteen years of spreadsheets already live.

So I gave it the right job. Instead of asking Cowork to build a tool and remember my farm for me, I pointed it at the farm. I connected the folder where my spreadsheets already sit and started asking my questions there. Now the answers come from my actual yields, my actual prices, my actual field notes. Not hardcoded numbers frozen in an artifact. And because the files stay put, nothing gets forgotten when I close the laptop. The spreadsheet holds the history. Cowork is the mind I point at it.

Cowork is a version of Claude that can build spreadsheets, pull live information off the web, work through multi step tasks, and connect to the tools you already use. You hand it a job and it comes back with the finished deliverable, a spreadsheet, a document, a marketing plan, that you can download and reuse. Play around with it. Ask it what it can do for you.

I’ve noticed that Claude is making it easier and easier to access the functions of Cowork. It’s pretty simple. Open the Claude desktop app on your computer and start a new chat. In the latest version of claude, you just toggle on “cowork.” Find the option to add a folder and point it at the place your farm data already lives. As you can imagine, the more data that you have and the better it’s organized, the better the responses will be. Claude reads across every file in that folder and answers from your actual numbers, not a guess. Save it as a Project so it remembers you the next time you sit down.

Cowork is also incredible at taking messy data, and putting it into spreadsheets. Drop that incredibly long purchase invoice into cowork and ask it to create a spreadsheet for you. Then save that spreadsheet and put it in the right folder. The next time that you ask a question across that folder, claude has the new context to go along with the old. We’re slowly building a system.

We’re starting to build tunnels. We are slowly connecting our data, and our context, to the AI so that it can do meaningful work for us. Next week we will explore the data we don’t have, and how we can close that gap.

One bit of housekeeping. I’m turning on paid subscriptions this week. I have a feeling that the Substack algorithm rewards the writers who make them money. Just a hunch. My writings will continue to be free, but I just wanted to give a heads up to the ones that have pledged support. Thank you.

One more thing. Starting Thursday, August 6th, I’m starting a second, smaller series about living off the land. These will be much shorter posts and have nothing to do with AI. The posts will cover topics from gardening, to hunting, to fishing, to trapping, to recipes. My wife Cali will be a regular guest writer. It will be a look inside the life of two delta farmers. The AI and farming series will still be my main focus, but I can’t wait to share my gumbo recipe with you. It’ll knock your socks off. Thanks for following along.

Sources and further reading:

The Claude Cowork Artifact

Prospective Plantings Spreadsheet

Subscribe now

Share

His name was John Snow. Cholera had broken out in Soho, a crowded neighborhood packed around a street called Broad Street. It killed fast. Something like six hundred people died in ten days, most of them within a few hundred yards of each other. The whole city believed cholera traveled through bad air. A foul smell, a vapor rising off the river, something you breathed. That was the settled science of the day, and it was wrong.

Snow did not believe the air theory. But believing is not proving, and nobody was going to shut down a neighborhood on one doctor’s hunch. So he did something almost nobody had thought to do. He went and got the record.

The deaths were already written down. Names, addresses, dates. That information existed before Snow ever showed up. It sat in registers and parish books, scattered, unread, useless. Snow took a plain map of the neighborhood and he made a mark for every death, right on the house where it happened. And when he was done, the map answered a question no smell and no theory ever could. The deaths clustered around one thing. A public water pump on Broad Street. The people who drank from it died. The workers at the nearby brewery, who drank beer all day and never touched the pump, lived. The men and women who had their own well, lived.

Snow took that map to the parish authorities and got them to pull the handle off the pump. The outbreak was already fading by then, and honest historians will tell you the handle came off near the end, not the start. But that is not why we remember it. We remember it because a man took scattered facts that were already sitting there, gave them a shape, and the shape pointed straight at the answer.

What was missing was not data. What was missing was the structure and the right question. The same registers that had been useless for weeks became the founding document of modern epidemiology the moment somebody organized the register and asked the right questions. The next time you are in London you can grab a pint at the John Snow pub and look at some of his work.

Last week you started becoming the person who records the facts. This week is about the harder truth nobody tells you. The facts alone are worthless. It is the shape you put them in that is worth keeping.

A pile of data is not a system

Last week we started the boring work of capture. You started recording data on purpose and putting that data in folders that are yours. That is step one and it is the right step. But I need to be straight with you about something, because if I let you believe that a folder of files is the finish line, you will quit too early and never see the payoff.

A folder of files is a woodpile. A system is a different thing.

The difference between a junk drawer and a filing cabinet is not what is inside them. Both hold the same paper. The difference is that one has a shape and the other does not. The idea behind a filing cabinet is that it allows you to easily find the information that you are after. You can hand a filing cabinet to a stranger and they can use it. A junk drawer only works for the one person who filled it, and only until he or she forgets (I just got a flash of anxiety thinking about our junk drawer). A system is a pile that has been given a shape so that anybody, you in five years, your son, or a machine, can walk up to it and get an answer out.

That is what a system is. Capture, so the fact gets recorded. Structure, so it can be found. A way to ask, so it gives back answers. And a loop, so what you learn this year is sitting there waiting for you next year. Snow had all four. The deaths were captured. The map gave them structure. The clustering was the answer. And the method he invented became a loop that has saved millions of lives since. You are building the same four things on your farm, at a much smaller scale, for much lower stakes, but the shape is identical.

Why this is the most valuable thing you are not doing well

Your farm throws off a fortune in information every single season. Every rain, every spray, every thing that worked and every thing that did not, every observation you make from the seat of the tractor. It is real data and it is expensive to produce. You paid for it with a whole year of your life. And almost all of it evaporates. It leaves the same day it arrives. By the time you need it, it is gone, and you are guessing.

That is the whole problem, and it is bigger than it sounds. Think about what makes a business get smarter over time. It runs the same play over and over, learns what works, and keeps the lessons. A restaurant serves a thousand plates a week and finds out fast what sells. You do not get a thousand tries. You get about forty. Forty harvests in a working life, one shot a year, and then the season is over and you do not get that year back. There is no faster way to run the experiment. So the only way you ever get wiser is to squeeze every drop of learning out of each of those forty years and carry it forward.

That is the entire game. And you cannot carry it forward in your head, because memory does not work that way. Memory keeps the story, and skips most of the facts. It remembers the year the bottom flooded because it hurt, and it quietly loses the three quiet years that would have told you whether the flood was normal or rare. A record does not have feelings. It keeps all forty years the same, so that year forty can talk to year one. That is why the record is worth more than your memory. Not because your memory is bad, but because a record does not forget the boring years, and the boring years are where the truth usually hides.

Here is the second reason, and it is the one that matters most right now. In this exact moment in history, companies are racing to commoditize knowledge. Everybody is going to have access to the same powerful AI for less than a dollar a day and it will be the same for everyone. So think about what is left. If the tool is the same for everyone, the only thing that separates a useful answer from a generic one is what you feed it. And the one thing you can feed it that nobody else on earth has is the history of your own ground and business. Your fields, your rain, your yields, your overhead cost, your combined years of hard learning.

That is the one input that is yours and yours alone. An organized record of your farm is the thing that turns a generic tool that talks to everybody into an advisor that works for you. The farmer with an organized record is going to have an advantage in the AI era that the farmer with a shoebox of receipts simply cannot buy. The data is the moat. The tool is just the tool.

And here is the third reason, the quiet one. The quality of the question you are allowed to ask is capped by the quality of your record. A system lets you ask which of my fields actually made money, and which one only looked like it did. You cannot ask a question your data cannot answer. So every hour you spend giving your data a shape is not filing. It is raising the ceiling on the decisions you are able to make for the rest of your career. That is not clerical work. That is the highest leverage thing on the whole place, and it is sitting there disguised as a chore.

The step by step: connect your record to the chat.

So here is the actual skill this week, and it is smaller than the sermon around it. You are going to point AI at your own files and ask it a question. There are three ways to do it, from fastest to most durable. Start with the first one today. You do not need the other two to begin.

One. Drag the files straight into a chat. This is the two minute version and it needs nothing but the files you already have. Open your AI tool and start a new chat. Find the handful of files that are actually about what you want to ask. Not the whole farm. Just what is relevant. Drag them from your folder right onto the chat window and let go, or click the attach button, the little paper clip, and pick them. You will see them load into the message. Then type your question in plain English, the way you would ask an employee who had read all of it. That is the whole thing. As of now the chat reads plain text, Word, PDF, spreadsheets, and photos, which covers everything we built last week.

Two. Load it once into a Project so you stop re dragging. After you have done the drag and drop a few times you will get tired of hauling the same files into every new chat. That is your signal to build a Project. We talked about projects a few weeks ago, but essentially they are a workspace where you upload your files one time, write your standing instructions, and then every chat you start inside it can already see all of it. You are not re-explaining your farm every morning. On most platforms the Project feature sits on the paid tier, so this is the point where it may cost you something.

Three. Connect the folder so it reads your files live. If your records already live in something like Google Drive, you can connect that directly to your AI tool so it reads the folder without you uploading anything. It can read your files and save new ones, but it cannot edit, move, or delete a single thing you already have. Your record stays yours, untouched. And it always sees the latest version, so when you update a note, the tool sees the update next time you ask. This is the version where the woodpile and the reader finally stop being two separate chores. One caution. Every tool handles your data differently, so read the privacy terms for yourself before you connect, and think twice about pointing it at anything truly sensitive.

Do the first one today. One new chat, the few files that matter, one real question about your own ground. That is the whole assignment. The other two are just the same move with less friction.

I slowed things down this week on purpose. Last week was about starting to capture. This week I want to give you room to actually do it before we build the next piece on top of it.

Here is what I have been doing while you catch up. For the last few months I have been photographing my handwritten notes from seasons past and letting AI transcribe them into clean digital text. I filed away photos of my 2026 crop certification. I stored my rainfall data. I logged my irrigation events, field by field. Planting data. And I took my own journals from more than twenty years back and had AI turn those into files too, so a note I wrote by hand two decades ago can finally sit in the same system as this year’s rain.

None of it happened in one sitting. It happened a little at a time, a few notes here, a folder there. If you weren’t able to start last week, now is a great time to start. Because next week the bill comes due, and this time it pays you. Two issues of work with no payoff. That ends. Next week we start building the tools for our data.

Sources and Further Reading

John Snow - Historical Giant in Epidemiology

Obsidian CEO, "File over App"

I use the Johnny Decimal system to organize my files

Share

Subscribe now

Leave a comment

The building was the seed bank now known as the Vavilov Institute of Plant Industry. Inside it sat the largest collection of crop seeds in the world at that time, decades in the making. Beans. Corn. Wheat. Rice. Potatoes. Outside, the city was under siege and starving, and yet the men inside would not eat what they guarded. A rice specialist named Dmitri Ivanov died surrounded by thousands of bags of rice. Science History said that nine scientists died guarding the seeds.

To understand why, we look to the man the seed bank is named for, Nikolai Vavilov. He was a Russian botanist who, decades before anyone used the word biodiversity, understood that the world's food supply depended on protecting the full variety of its crops. Gary Paul Nabhan, who wrote a book retracing Vavilov's travels, said that if justice were done, Vavilov would be as famous as Darwin.

One thing drove his whole life. He wanted to end famine. He was born in Russia in 1887, and all through his childhood something caused crop failure every few years. Drought. Insects. Disease. One famine alone took four hundred thousand people. When the government set up relief kitchens, the food was barely food. Tolstoy called it “vomit, regurgitated by the rich.”

So Vavilov went at the problem from the root. He spent decades crossing five continents gathering seed, hoping that the disease resistance and drought tolerance of those seeds could stop the next famine before it started. Those seeds are what filled the building in Leningrad in 1941 and 1942. The seeds that were meant to save the world. Eat the seed and you eat the knowledge. Vavilov never saw how it ended, but I am certain he was proud of the men who gave their lives protecting his life’s work. He was arrested in 1940 and died of starvation in a Soviet prison in January 1943.

Some things are worth more preserved than something you consume today. And a collection like that does not survive by accident. Somebody decides it will survive, and then does the unglamorous work of protecting it.

That is this week’s whole subject. Last week I wrote about how the knowledge on your farm erodes, the same way topsoil erodes, when the person who carries it in his head leaves. This week is about the first job in fixing that. But first, a personal story about why this matters to me.

My grandfather never wrote it down.

I loved my grandfather very much. He gave me my first job at his convenience store, paid me when I made good grades, and taught me lessons that bring tears to my eyes to this day. On one occasion, I must’ve been seven or eight, I decided to steal a handful of Big Red and Juicy Fruit gum. I slid the packets into my pocket right before we closed up for the day. Little did I know, Pops had seen me. He confronted me, and said, “Son, you don’t need to steal from me. Everything I have is yours. All you need to do is ask.”

I often think back to that moment when I have the chance to offer grace to someone. Pops saw me in one of my worst moments and offered love instead of anger. He offered grace instead of wrath.

A few years later, he was murdered in that same store. There one day, gone the next. I don’t share that for sympathy. I share it because of my own desires and the role his death played in my life. I still desire to have a note or letter written by him to me. I still desire to know what it was like farming in the 80’s when he was a farm manager in the same area where I farm now. Was it harder then? Had the “Friars Point split” fully formed at that time (farmers in my area call it the Friars Point split because the Mississippi River always seems to break up the rain clouds right at Friars Point)? What would you tell your 38 year old self?

My grandfather didn’t farm the exact same ground I farm today, but he did manage the fields that border where I farm. If I had them, I would consider his notes and knowledge a gold mine of information. Luckily, I have my father-in-law Johnny, who has been on this ground his entire life to guide me through some of those knowledge gaps. He knows the ditches we missed. What ground will make cotton and what ground won’t. He also knows the history of this place much better than me.

I keep notes in my phone from my conversations with Johnny. In many ways, a lot of his knowledge has become mine. But a note buried in my phone is not a record, and a father-in-law who could remember everything is just one man. I learned what that costs the hard way with Pops. So I do not leave it to memory or to luck anymore. When Johnny tells me something about this ground, I write it down on purpose, and I put it somewhere that is mine to keep. That habit is the whole job this week, and it is simpler than you think. Let me show you how.

Own the file, not the app.

Here is the mistake almost everybody makes when they finally get organized. They go shopping for software. They ask which app, or which subscription is going to solve this problem. That is the wrong first question, and I want to talk you out of it before you spend a dollar. The thing that actually lasts on your farm is not an app. It is a file. A plain document. A recording of your father explaining how much a rotation means for your yields. Apps come and go. Companies get bought. Subscriptions lapse. You switch from green to red. But a folder of plain files that you own, sitting on your own computer and backed up, will outlive every one of them.

You already know this in another form. Your data is likely scattered all over the place right now. John Deere Operations Center, Climate FieldView, some in your accountant’s software, some in your grain buyer’s portal, some in a few notebooks, and a whole lot in your own memory. Every one of those platforms will tell you that you own your data and can export it whenever you want. But owning your data inside somebody else’s platform is not the same as having it. If your whole record lives inside six companies’ websites, then it’s really six companies’ records that you are allowed to look at. The day you switch equipment colors, or a company gets sold, or a login stops working, you find out how much of it you actually kept.

Before you organize anything, before you touch any AI tool, you build one place that is yours. A farmer-owned file layer. Everything feeds into it. That is capture.

The step by step.

Let me walk through actually making one of these files, because this is where most people give up. We are doing something very simple. We are writing something down and putting it in the right folder with a date on it. That is the entire skill.

On a Windows computer. Open the folder where the note belongs, say the folder for the Homeplace. Right click in the empty white space. Choose new document, then Text Document. It appears with the name highlighted, so type the name right then: 2026-07-14_Homeplace_notes and hit Enter. Double click it, type what you know, and hit Save. Done.

On a Mac. Open the program called TextEdit. It came with the computer. Before you type, click Format at the top of the screen and choose Make Plain Text, one time. Then type your note, hit Save, give it the date-first name, and pick the folder it belongs in. Done.

The phone has one job in this system, and it is capture. Take the picture of the receipt. Record the voice memo in the tractor. Type the quick note in your Notes app. Then once a week, sit down at the computer for ten minutes and move what the phone caught into the right folders. Email the photos to yourself if your phone doesn’t connect to your computer, or drop them in the cloud folder—however you already move pictures around. The phone catches it. The computer files it.

And if you already have a file that is named wrong, fixing it takes five seconds. Right click it and choose Rename. On a Mac, click the name once and pause. Put the date first, then the subject.

Here is the whole loop, start to finish. It rained three inches Tuesday night and the low spot on the north end held water until Thursday. The ditch reverse flowed into the field and flooded the bottom 20 acres. You open the Homeplace folder, make a new text file, name it 2026-07-16_Homeplace_rain, type three sentences, and save. Sixty seconds.

Do the brain dump interview first, because that knowledge is the seed that is actively leaving.

Sit down with the oldest person on your operation. It could be your father, grandfather, it could be an employee who has been there forty years, or it could be you. Record it on your phone. Ask the questions that are only in somebody’s head. Which fields drain slow, how slow, and why? What fields do better in wet years? What would you change about the farm? What do you think would be the best investment for the farm? Then get that recording turned into text. Every phone transcribes voice now, and any AI tool will turn a recording into clean text in minutes.

I would give most anything for two hours with my grandfather behind that counter, just talking about the farm and life. Just the thought of it takes me back to Keanu Reeves on Colbert when Colbert asked, “What do you think happens when we die?” Reeves took a breath, and said, “I know that the ones who love us will miss us.”

We do.

Photograph and scan the paper notes and letters before they are gone. The shop notebook. The hand-drawn field maps. The rain gauge log taped inside the shop door. Snap a photo of every page with your phone.

Name your files so a stranger could understand them. This is the boring step that makes everything later possible. Put the date first, year then month then day, so files sort themselves. Then the field or subject. So a file becomes something like 2026-07-14_Homeplace_notes. You will thank yourself later, and so will any tool you ever point at these files.

Put it all in one folder tree that is yours. Not scattered across your desktop. One top folder, call it whatever you want, with a handful of subfolders inside: Fields, Financials, Leases, Varieties, Equipment, People and Vendors, and a Knowledge Base for those interviews. I have built a starter version of exactly this folder tree, empty and ready to fill, CLICK HERE if you want to try it out. There are some instructions inside the folders.

If you want to start today, tap click here and download the file. Then tap the file itself. Your phone or computer will unpack it into a folder called Farm_Files. Move that folder somewhere permanent. Not your Downloads folder, which is where files go to die. Open START_HERE and read it. It takes two minutes. On a Windows computer, right click the file and choose Extract All first. Everything else works the same.

The honest limits.

I will not oversell this. Capture is the least fun part, and it is real work. The brain dump might take two hours, and turning shop notebooks into files is a rainy afternoon job you will probably put off. So let me be honest about three things.

First, this week does not give you a payoff. You are building the woodpile before the cold snap. A folder of well-named files, by itself, will not tell you which variety to plant. That comes later in this series, and I will not pretend otherwise.

Second, you do not have to do it all at once, and you shouldn’t try. Capture the knowledge that is most at risk first, meaning whatever lives in the head of the oldest person on the place. The rest can come field by field over a season.

Third, this is deliberately vendor-neutral. I have not told you to buy anything or which AI to use, on purpose. If your files are yours and in a plain format, you can feed them into any AI tool that exists now or later. That is the whole point of building the file layer first.

A folder full of files is not the same as a system. What you will have is a very well-organized woodpile. It will not stack itself, and it will not answer a question when you ask one.

Next week I will show you how to turn that pile into something that remembers, something you can actually ask. We’re getting to the fun stuff.

Leave a comment

Sources and further reading

An article about Vavilov and his team

Your starter "filing cabinet"

Gary Nabhan's book about Vavilov

Subscribe now

Share

But I had two helpers last week. My 11 year old nephew and my 8 year old daughter. My nephew loves the farm, and he has a work ethic you can’t teach. We took turns driving the tractor and shoveling dirt on top of the poly pipe to keep it in place. After my nephew left, my daughter helped out. She poked a few holes in the poly pipe and drove my truck along behind me so I didn’t have to walk the quarter or half mile back to the truck. Eight years old, creeping down a turn row, proud as she could be. I was thankful for both of them.

It’s one thing to think about the next generation and what opportunities may come their way. It’s another thing to watch them work. I couldn’t help but think back to my first years on the farm and how fast things have changed. We no longer have chopping crews walking the rows with hoes, killing the weeds that became resistant to Roundup. The boll buggy doesn’t run cotton from the middle of the field to the end anymore. The module builder on the farm is now a chicken coop.

But as I watched the kids work this past week, I kept coming back to one question. What exactly are we hoping to pass down?

The easy answer is the soil. We work at that every year, and I’ve written about it here before. But there is a second inheritance on every farm, and it erodes faster than topsoil. The knowledge.

Think about what actually runs your operation. Which field drains slow after a two inch rain. Which variety burned you and why. The handshake agreement with the landlord. What the soil test said, what you did about it, and whether it worked. Almost none of that is written down. The part that is sits scattered across a notebook in the shop, a spreadsheet on an old laptop, and an Ops Center login nobody else has. The rest lives in exactly one place, somebody’s head. Topsoil takes decades to erode. This kind of inheritance can disappear in a single funeral.

My daughters may want to farm this land someday. They may not. Either way, whoever farms it next should not have to reconstruct what I know from memory and scattered paper. The soil and the knowledge ought to be passed down together, and that only happens if the data is in a place that’s accessible.

That is the problem I was chewing on from the tractor seat, and it turns out we’ve been building toward the answer for five weeks without calling it that.

For five weeks we have been building one thing. We gave the AI context with the ninety second habit. We made that context permanent with a project. Last week we wrote it a job description and put it to work as an employee you supervise.

Today we put that employee on the morning shift.

The market moves before you are out of bed. A rule changes in Washington and you read about it three weeks later in a magazine. A frost shows up in a forecast you did not check. Cotton gaps up overnight and you find out at the gin. The information that should change your decision is out there every morning, scattered across a dozen places you do not have time to check, and by the time it reaches you the moment to act is gone.

You set it up one time. From then on, every morning before you pour your coffee, your employee reads the markets, the weather, Washington, and your inbox, and leaves a one page brief on the kitchen table. That is the daily convenience. But something bigger is happening underneath it. Every brief gets written down and kept, and day by day those briefs stack into a running record of what the farm knew and when it knew it. This is the first entry in building our knowledge base. The farm starts writing things down.

Creating your first autonomous workflow

The brief is one job, written one time.

We could paste the job description below into Claude or ChatGPT every morning and the brief would work. It also costs you two minutes and the discipline to remember, and you already know from a few weeks back what happens to a habit you have to do by hand every single day. You quit.

There is a part of Claude called Cowork that lives in the desktop app, the one you install on your computer, not the website. It can run a job on a timer instead of waiting on you. You write the job once, tell it when to run, and it runs itself every morning from then on. Cowork is on the paid plans, so this is the first thing in this whole series that asks you to spend a little money. I will be straight with you about that. If free is where you want to stay, run the job by hand each morning. The brief itself is the same either way.

Step one: give Claude your inbox

Half of what runs a farm shows up in the inbox. The broker’s morning comment. The quote the fertilizer dealer sent last night. The notice from the FSA office. The landlord asking about the ditch. If the AI cannot see your email, it cannot flag any of it, and you stay your own filing clerk.

Connecting takes about a minute:

1. Open Claude and go to Settings, then Connectors. In the desktop app, look for Connectors under Customize in the sidebar. On the phone, you can also tap the plus button in the message box and choose Connectors.

2. Find Gmail and hit Connect.

3. Sign in to your Google account and approve the access it asks for.

That is the whole setup. From then on, when you ask Claude a question that touches your email, it searches the inbox and reads the threads that matter. Ask it what the elevator said about basis last week and it finds the actual email instead of you scrolling for it.

Two honest notes, because you deserve them. First, Claude can read your email and it can write drafts, but it cannot send. Every reply it writes sits in your Gmail drafts folder until you look it over and hit send yourself. I consider that a feature, not a shortcoming. Nothing goes out with your farm’s name on it until your eyes have been on it. Second, on privacy: Claude only goes into your email when you ask it something that requires it, it can only see what your Google account can see, and Anthropic states plainly that it does not train its models on your Gmail data. If you use ChatGPT, it has its own Gmail connector in its settings and the concept is identical. The walkthrough here is Claude, because Claude is where we are going to schedule the job.

Step two: schedule the job

Open the Claude desktop app and go to Cowork. Start a new task. Type /schedule in the box. Claude will ask you a few questions about what the job is and how often to run it. Pick daily, or weekdays if you would rather skip Sunday. Pick a time a little before you usually sit down, so the brief is waiting when you get there. Then hand it the job below.

THE MORNING BRIEF, LONG LAKE FARMS

You are the morning analyst for Long Lake Farms, a 3,000 acre row crop operation near Friars Point in Coahoma County, Mississippi. We grow cotton, corn, soybeans, and wheat. Every morning, build me a short brief on what changed overnight that could impact my decisions. Cover these five sections, in this order.

Markets. The overnight and prior day move in corn, soybean, cotton, and wheat futures. Note any move worth my attention, and why it happened if a reason is reported. Include anything notable in diesel, fertilizer, or the dollar.

Policy. Anything in the last day on the Farm Bill, USDA and FSA program notices, crop insurance and RMA, EPA label or pesticide actions, and trade or tariffs that touch US crop exports.

Weather. Today’s forecast for Coahoma County, and anything in the next seven days that threatens field work or the crop. Flag rain, heat, or frost plainly.

My crops right now. Anything tied to cotton, corn, soybeans, or wheat at the current growth stage in the Mid South that I should be thinking about this week.

My inbox. Search my email from the last 24 hours. Flag anything from my broker, the gin, the elevator, my input suppliers, my landlords, FSA, or my insurance agent that needs a reply or a decision from me. One line per item, with the sender and what they need. Skip the junk mail and the newsletters.

How to write it. Keep the whole thing short enough to read over one cup of coffee. For every claim, give me the source and a link. Lead with what changed and what, if anything, I should do about it. If nothing material happened in a section, say so in one line and move on. Do not pad.

Two hard rules. First, never give me a price or a number as fact unless you have it from a source and you link it. If you are not certain of a figure, say so and tell me where to check it. I will treat every price in this brief as something I verify against my own market source before I act on it. Second, append today’s brief to a running file called morning brief log, so tomorrow you can see what you told me yesterday.

Swap your operation in for mine. Your acreage, your county, your crops. If you run cattle or grow rice, add a section. If you are not ready to let the AI into your email yet, delete the inbox section and the brief still works. The shape of the job does not change.

Cowork only runs while your computer is awake and the desktop app is open. If your laptop is shut at five in the morning, the brief does not run until you open it back up. For a morning brief that is fine. You open the computer at six, it runs, and the brief is there by the time the coffee has finished brewing. Just know that is how it works.

Before you trust it, run it one time by hand and read what it gives you. Is it pulling the right things? Too much? Too little? Did it flag the emails that actually mattered? Tighten the job until the first brief reads like something you would actually sit with. That is the only real work. After that it runs itself.

The brief tells you where to look. It does not tell you the number.

Use it to know that cotton moved overnight and roughly why, so you walk into the day pointed at the right question. You still verify every price against your own source, your DTN, your broker, your screen, before a dime moves on it. Same with the inbox. The brief tells you the gin emailed about the seed rebate. You read the email yourself before you answer. The brief is a scout, not a signature.

This week’s job

Connect your email and build one morning brief. Do not try to cover everything on day one. Markets, one or two policy scans, and the inbox check is plenty. Run it once by hand, read it, tighten it, then put it on the timer.

And understand what you just did, because it is bigger than a convenience. Every morning that brief writes itself into the log. By Christmas, a season’s worth of markets, weather, and decisions will be sitting in one file instead of scattered across your memory and mine. That file does not care if you sell the farm, hand it to your daughter, or get hit by a truck. It persists. That is the whole point of this series. The farm has started writing things down.

But notice what the log holds and what it does not. The brief captures what the world told the farm. Nobody is capturing what the farm told you back. Which fields drained slow. What the variety did on your dirt. Why you made the call you made. Right now that knowledge has one copy, and it's you. Next week we start making a backup.

Sources and further reading

Schedule recurring tasks in Claude Cowork

Connecting Gmail to your claude account

NASA had the blueprints and still lost the rocket. Here's the story.

Subscribe now

Share

Leave a comment

Maybe he’s right. Maybe he’s not. Color me skeptical

And here we stand, a few days before we will celebrate 250 years of self governance, divided about what this machine should be. I won't pretend to know how the story ends. Elon doesn’t know how the story ends. I am also not naive about the cost of it either. These models get built on real ground, with real water, power, and steel, often in the same rural places we farm, and that is a conversation we should be having out loud, not after the fact. There are real questions too about whether a tool this powerful gets built safely, and whether it answers to the people or just to the people who own it. The promise of abundance is hollow if it only ever reaches the businesses big enough to chase it.

I am not going to stand here and tell you that I have squared all of this in my mind. We spray chemicals and use synthetic fertilizers on our farm. We plant cover crops, pollinator plots, and do no-till when the penetrometer allows. I live inside the tension already. This tool is one more version of it, a real cost I cannot fully measure yet, set against better decisions I can see on my own farm. I am not asking you to pretend the cost away. I am asking that we keep our eyes open, take honest inventory, and demand the people building this do it with care. That is not a settled answer. It’s the only honest one I have. I suppose it will take some time to work through the cost side of that ledger.

I do sincerely hope Elon is right, and that the decades ahead are better than the ones behind. We have already watched these tools chip away at hard problems, predicting the shapes proteins fold into, a problem that stumped biologists for fifty years, and speeding up the science that heals people. That is real, and it is reason enough to pay attention. But out here, the work is more humble, I suppose. Nobody is getting healed on the turn rows of Long Lake Farms. But I do find myself stress testing my own gut against the machine. Is this the right application at the right time? Is muriate of potash the best potassium choice for my ground? What impact will this application have on my soil?

But I can hear the doubt in my own head, questioning if the tool really sharpens every one of those decisions. Does it actually get me anywhere? To answer that straight, I have to go back to what I told you last week. Willard Cochrane, and the treadmill. A new tool shows up. The first farmers to buy it make money. Everyone else buys it because they see the benefits proven, production climbs, the price of the commodity falls because of increased production, and the gain you paid for gets competed away. What I did not say plainly last week is this: the tool itself is priced to take its cut first. And it should.

For example, John Deere does not sell their cotton picker at the cost of the steel. They price it to capture most of the efficiency it hands you. The round bale picker eliminated three jobs on our farm the day we bought the picker. We also sold the boll buggy and module builder not long after the picker was on the lot. Three people, two pieces of machinery. No more stopping to dump cotton on the boll buggy. No more scrambling for seasonal employees. No more picking up cotton off the ground. But as wonderful as that tool is, it’s not a magic wand to profitability.

Almost all tools on the treadmill are priced to capture most of the value they create. The picker's price climbed with every hour of labor it saved you. That is the rule. The bill scales with the benefit. I don't believe this tool follows the same pattern. Your subscription is a fixed cost, and a small one, and it does not climb the year you nail your nitrogen or keep a variety off the wrong ground. For the first time, the cost of the tool is cut loose from the value of what it does. So the math runs backward from everything we are used to. One better marketing call. One input pass you skip because it was never going to pay. One variety you keep off the wrong ground. Any one of those can return more than you will pay for the tool in a decade.

Does that beat the treadmill? Probably not the way we want it to. If every farmer makes better decisions, the price still falls, and Cochrane still wins in the aggregate. But there is a real difference between a tool that costs about what it returns and a tool that costs almost nothing and might return a lot. The treadmill does not stop. It just gets cheaper to run on. And the person who runs it cheapest, who spends the least on the things that do not pay and gets the most out of the decisions that do, will be better positioned in this new era. At least that’s my theory.

In previous episodes, we discussed how to give AI the context it needs to provide better answers. We set up your settings and created your first project. Today, we are going to set up your first employee.

Why an employee, not a robot or a brain

A few paragraphs back I told you the picker took three employees off this farm. Here is the part that still sits with me. The tool we will set up today runs the other direction. It hands you a worker back. Not a hand for the turn row, this one will never grease a combine or fix a flat. But for the work that piles up in the office, the reading, the figuring, the drafting, the numbers you never have time to chase down, this tool can help.

Do not mistake it for a robot or a brain. A robot replaces your hands. A brain replaces your judgment. This is neither. It is a “helper” you give jobs to and check behind, and you stay the boss. We farmers are skeptical people. We would not hand a new man the keys to the spray rig on day one. You start him small, you check his work, you give him more rope once he earns it.

Sharpen the job description

Two weeks ago you built your project and gave it a rough intro, like you would an agronomist on his first day. I also hope that you have started to add the documents that will give more context about your farm or business. Your soil maps, the notes from your agronomist, the research you’ve done on your farm, the research you trust, irrigation manuals, insect guides, etc. This week, we turn our rough intro into a real job description. Same project, same box. We are not starting over. We are swapping two loose paragraphs for four tight parts. I made this so you can copy and paste it into your project.

EMPLOYEE HANDBOOK [FARM NAME]

WHO YOU WORK FOR

You are an office assistant for [Farm Name], a [acreage] acre

[crops] operation near [town], in [county], [state].

WHAT YOU NEED TO KNOW

Acreage and crops: [e.g. 800 corn, 900 soybeans, 700 cotton]

Equipment: [planters, sprayer, combines, irrigation]

Team: [who works here and what they do]

Fiscal year: [when your operating year runs]

Key people: [landlord, lender, crop insurance agent, seed dealer, cooperative, grain buyer]

YOUR JOB, FOR NOW

Summarize long documents and emails.

Explain confusing paperwork in plain English. Answer questions about my operation using the files in this project. Bigger jobs come once you earn them.

HOW I WANT YOU TO WORK

Ask a clarifying question when you are missing something, instead of guessing. Use plain language. When you give me a number, show your work. If you are not sure, say so plainly.

That’s it.

The first day on the job

A good manager does not lecture the new hire. He hands him something and watches. Three small jobs, easiest first. Every one is safe. Every one you can check in ten seconds.

One. Introduce yourself. Find out if he read the handbook.

Based on what you know about my farm, what jobs can you help me with? List them.

A good answer names your operation back to you. Your crops, your county, your files. If it talks in general about farming, your handbook is too thin. Go add to it.

Two. Decode something. Find a paper you have read three times and still do not fully understand. A clause in a lease. A letter from FSA. A page of your insurance papers. Paste it in.

Explain this in plain English. What does it actually mean for me?

This is where it earns its keep, and it never leaves your screen. You are not asking it to speak for you. You are asking it to read for you.

Three. Summarize something. Find the longest, most boring email in your inbox. A seed pitch. A program notice. Paste the whole thing in.

Summarize this in three bullets, and tell me if there is anything in here I actually need to act on.

You stop reading the long version. You read a summary and a verdict.

Read what it gave you and ask, is this right? Just a quick reminder, this is a tool to help us make better decisions, not make decisions for us.

The two rules

Rule one. Give it jobs where a mistake is cheap. A wrong summary costs you ten seconds. A wrong spray costs you a field. Notice what we did not do this week. We did not ask about nitrogen rates. We did not ask about a spray call. Those jobs are coming. They are not first day jobs.

Rule two. Check the work until he earns the right to be unchecked. The first few weeks you read everything behind him. Once he proves a job, you ease off that job. You never hand over all the keys at once.

This week’s job

Open your project. Replace your instructions with the handbook. Give your new employee a new job. Then hit reply and tell me which one surprised you. I read every one.

Next week, we put him on the morning shift. Before you pour your first cup of coffee, he will have read what’s happening in Washington, the markets, the weather, and he will have left a brief on your screen.

A great report on AI and Energy

Elon on Moonshots-Apple Podcasts- (1:14:20)

Share

Leave a comment

I am guessing that is not what you expected to read when you clicked on this, but here we are. Thanks to the kindness so many of you have shown me over the past few weeks, I decided to sit down and write a proper introduction for those of you who do not know me. My name is Mark Noland, and I live 11 miles north of Clarksdale Mississippi near a small community known as Friars Point. I live on the farm with my wife, Cali, and our three daughters, Lilla, Kate, and Emmie. We love this place.

I grew up in Dublin, Mississippi. I guess there is something special about the small towns of Coahoma County, and yes, we had a St. Patrick’s Day parade every year for the entire population of Dublin. All 43 of us. My parents still live there, and it will always hold a special place in my heart. Dublin is where I got my first job, restocking shelves and running the register at my grandfather’s convenience store. It is also where I fell in love with fishing and the outdoors. I spent whole summers on Dublin Lake in my twelve foot Jon boat, fishing for bream and bass. It was a wonderful childhood.

I played sports in high school, did well enough in school, and graduated from the University of Mississippi in 2010 with a degree in managerial finance. I think I chose that degree because of a love for math that started behind the register at Doug’s Quick Stop. For those of you old enough to remember, the recession of 2008 had gutted the job market, and by 2010 there were not many financial institutions in the south hiring a brand new finance graduate out of north Mississippi. So I took a job in insurance in Clarksdale, married my sweet wife, bought a house, and figured we were off to the races.

It did not take me long to realize I did not want to sell insurance for the rest of my life. Knowing I wanted out, I asked my father in law if I could help on his farm until I found something else to do. I quit the job at 9 AM on a Monday, and before noon I was on a tractor. I have been here ever since. The harvest this fall will be my sixteenth.

A New Chapter

We moved to the farm in 2019 when Emmie was born. Then Covid came, and my wife went crazy. Not Covid crazy. Not three kids messing with your hormones crazy. Not stuck in the house crazy. Crazy. It turned out there was a reason for her odd behavior, and the reason was black mold living under our house. She had been carrying chronic inflammation through her body for months without either of us knowing it.

During that time, getting well became a full time project for Cali. She read everything she could find on medicine and recovery, and somewhere along the way she stumbled into regenerative agriculture and started bringing it to the dinner table. We began talking about what it might look like on our own ground. In 2021 we planted our first cover crop, and we have been experimenting on the farm ever since.

Now let me come back to where I started, because when I tell you farming sucks right now, I am not whining. I am reading the same spreadsheet every farmer in this country is staring at. It costs a farmer about five dollars to grow a bushel of corn where the market will only pay him a little over four. Soybeans and cotton lose money for most growers across the country today too. Farm debt is on track to hit a record, nearly 625 billion dollars. Farm bankruptcies jumped 46 percent last year, and the state that filed the most bankruptcy claims was Arkansas, right across the river from where I am sitting.

An economist named Willard Cochrane described the whole thing back in 1958 and called it the agricultural treadmill. It works like this. A new technology shows up. Better seed, a bigger planter, a stronger chemistry. The first farmers to adopt it make good money for a season or two. Then everyone else adopts it, because they have to, and as the new technology spreads, production climbs and the price falls. The gain that used to be profit gets competed away, the value of the improvement quietly capitalizes into the price of land and the cost of rent, and the farmer is left running faster just to stand in the same place. The people who reliably win are the ones selling the inputs and the ones who own the dirt. The farmer in the middle keeps running. That treadmill is the real engine behind get big or get out, and it is why the number of farms in America recently slipped below two million for the first time in almost two centuries.

And while we have been running on that treadmill, we have been spending down the one asset we cannot reorder from the dealership. A study published in 2021 found that roughly a third of the Corn Belt has lost its topsoil entirely, the dark soil that made the ground worth farming in the first place. Closer to home, the Mississippi river that runs past Friars Point carries the nitrogen that we and everyone upstream of us pour onto our fields all the way down to the Gulf, where every summer it feeds a dead zone about the size of a small state. The weeds have caught up with us too. The pigweed that Roundup once wiped out now shrugs it off across most of the country, and in some fields it has learned to survive nearly everything we throw at it. We’ve been farming like the soil is free and the chemistries are permanent. It turns out that neither are true.

So why are we still doing this?

Cali’s family started farming this land about a hundred years ago. We still have the maps they drew by hand in the 1930s, showing what they planted and where. The farm kept animals back then, and a good share of the crops were grown to feed them. One of our long term dreams is to see a small piece of this place restored to that time. A time before synthetic fertilizer. A time before conventional agriculture was even a phrase.

That is the dream on one end. The reason I am still here, and the reason I am starting this newsletter, is that for the first time in sixteen harvests I think there may be a way off the treadmill, and it is coming from two directions at once. One is old and one is new.

The old one is the soil itself. It really begins and ends there. It seems obvious, and I promise if you asked one hundred farmers if they would like healthier soils, they would all say yes. But, the system doesn’t allow for health. The system is built for now, and now we have to chase yield so we can hopefully pay the bills at the end of the year. The system itself is built for exploitation, and never asks the hard questions like: what happens when my organic matter goes to zero? What happens when there is no top soil left? What happens if we keep polluting our streams, waterways, and oceans? On our farm, we are trying to do everything we can to protect our soil, but I’ll be honest, its incredibly hard to do while you are running on a treadmill.

The new direction is the one that genuinely surprised me, and it is the reason this newsletter exists. For my entire career, every new tool got sold to me as a way to grow more. More bushels, longer residuals, more acres to cover. The treadmill ate every bit of it. What is finally beginning to change is that the best of the new technology, and artificial intelligence in particular, is aimed at the other side of the ledger. It is not about growing more corn. It is about making better decisions with what I already have. A farm runs on a thousand small decisions every year, and the information we need to make those decisions is scattered everywhere it cannot help us. Soil tests in a binder, yield maps on a thumb drive, market prices on a screen, input quotes buried in a stack of emails, and the university research that should tie it all together sitting in PDFs nobody has time to read. For the first time in my life I have a tool that can hold all of that at once. I can sit at the kitchen table and ask it, in plain English, whether the potash rate actually fits what my soil test says, or what the trial data out of Mississippi State suggests for a variety on my ground, or what a five cent move in basis does to the budget on a particular field. It reads the research so I can act on it, and it does the analysis I used to either pay for or skip. That is the first crack of daylight I have seen in a long time, and learning to use these tools well, honestly and without the hype, is what deltafarmer is about.

So that is the dream on one end, an old farm pointed back toward the way it was, when we took what the soil gave us. On the other end is a screen full of data and a set of tools Cali’s great grandfather could never have imagined. This newsletter is where I am going to try to hold both of those things in my hands at the same time and figure out, in public and out loud, how a family farm in the Delta makes it through the next hundred years. I won’t be around to see that, but I do have a small role to play as a steward of this land.

I am glad you are here. Pull up a chair. Next week we will create your first AI employee.

This is one of my favorite books about agriculture: The Unsettling of America by Wendell Berry

An article about the pulse of US agriculture: Article Here

Subscribe now

Share

The trouble was the plow. Settlers had hauled their plows out from back East, cast iron bottoms that had cut the light sandy ground of New England just fine. In the black prairie gumbo, those same plows were near worthless. The sticky soil clung to the moldboard and would not turn it loose. A man would cut a few feet of furrow, stop, climb down, scrape the mud of his plow, then cut a few more feet and do it all over again. All day. Every pass. The tool worked everywhere on earth except the one spot that farmer happened to be standing.

Eventually, the man had repaired enough of those plows to know the problem cold. So one day at a sawmill, he spotted a broken steel saw blade, carried it home, and went to work. He cut the teeth off it, shaped it over a log, and polished the face until it shined. That polished steel did the one thing cast iron never could. It made the soil slide right off it. He called it self scouring. The farmers of the area called it the plow that finally worked, because you did not have to stop and clean it. It shed the dirt on its own, every foot of every pass. That mans name was John Deere.

The catch nobody warns you about

Last week I told you about the 90 second habit, and how to write standing instructions for your AI. Before you ask it anything, spend a minute and a half handing it context. Soil type. Crop stage. What you have already done. What you are trying to decide. Then we gave Claude and ChatGPT a set of instructions about who we are and how we want them to answer.

That habit alone will move your hit rate from generic garbage to something you can actually use. But I owe you the honest catch, because I walked straight into it myself. You have to do it every single time.

Every new conversation, you are retyping the same dirt description, the same county, the same rotation. After about a week of that, you start cutting corners, and going crazy. The context gets thinner (and Mark gets more mad). The answers slide back toward useless. And right about there, most folks quit (and I did too for a while).

That is the farmer climbing down off the seat to scrape the moldboard plow every few feet. It works. It is also exactly why nobody sticks with it.

There is a better way, and it only takes about fifteen minutes to set up. You do it one time, and from then on the AI shows up to every conversation already knowing your farm or your business. It is called a Project, and both Claude and ChatGPT have it.

A Project is Deere’s plow. Instead of scraping the same context off by hand every time you sit down, you shape the tool to your dirt once, and it scours itself from then on.

What is a Project and how do I get one?

A project is the space where you can store persistent context about your farm or business. It’s essentially the brain of your operation. To add a project in Claude or ChatGPT, tap the lines in the top left corner of your screen (iPhone). Then hit projects, new project. Name your Project. In each individual project, you have the ability to set instructions for your farm or business. In that section, tell it about your business, what you do, where you are located, your strengths, your weaknesses, and any other relevant information. After you have completed that step, upload any files (PDFs, DOCX, CSV, TXT, HTML, etc.) that are relevant to your work.

This is where projects begin to shine. Inside of my “Long Lake Farms” project, I have thousands of pages of context that is relevant to my farm. I have a digital copy of the Delta Agricultural Digest. The Delta Agricultural Digest, for those unfamiliar, is an annually published reference manual by Farm Press covering crop production practices, seed variety trial results, pest management rates, and other recommendations for major crops grown across the Mid-South farming region. That one document alone is getting me closer to my corner of the curve.

I also uploaded my soil test results. I input the Mississippi State Extension Irrigation Manual. I added the MSU Extension Weed Management publication. I also added the Insect Control guide from Mississippi State. I would highly encourage you to think about the documents that you consistently search, and add those first. And a shout out to the extension specialists and researchers across our region, who put out first rate work year after year. Thank you!

Now, we’re cooking with some hot grease

The visual above is not a comprehensive list of all the files that I have in my “Long Lake Farms” project, but it’s a great place to start. Hopefully, you are already uploading files into your own project and understand the concept clearly. From now on, every time that you ask a question inside of your project, the AI is reading the relevant files before it gives you an answer.

Let’s say you have a question about stinkbugs in soybeans. Instead of giving you the average answer across the United States, you are getting recommendations from the experts who live and work in your context. No more getting the plow right, it is built for your operation.

Below is a study about preoperative instructions at a hospital in Singapore that illustrates the power of projects and giving AI context that persists. Once hospital guidelines were loaded as context, the model improved its performance and performed 11.3 percentage points better than when no context was loaded.

The fifteen minutes you spend building your project is the last time you start from zero.

Every conversation after that, the AI shows up already knowing your county, your rotation, your soil, and what you are trying to get done this season. All you have to do, is click your project, and chat with the LLM just like you always have. Except now, you aren’t an anonymous farmer asking a generic question anymore- you are “Long Lake Farms,” and the tool knows the difference.

Your one task this week: build your first project. Name it after your operation. Write two paragraphs of custom instructions like you are introducing a new agronomist to your farm on his first day. Upload whatever you have sitting in your email and what you can find for free on the web: your latest soil tests, your APH summary, your FSA-578, the Delta Ag Digest, the Insect Control Guide. It does not need to be perfect. It needs to exist.

Next week I will show you what happens when you put projects to work

Subscribe now

Leave a comment

To understand the cockpit, we’ve got to go back in time to the 1920’s. At that time, the United States Army (the Air Force wasn’t created until 1947) set out on a mission to build the cockpit. They measured the physical dimensions of hundreds of pilots to get an “average” size, so they could build the cockpit around those specifications. Seems reasonable, right?

In 1950, after the United States realized that they had a major problem, a 23-year-old lieutenant named Gilbert Daniels was tasked with researching cockpit design for the Air Force. Two years later, in 1952, Daniels published his report named “The Average Man.” He measured 4,063 pilots across 140 body dimensions and built an "average pilot" from the ten dimensions that mattered most for cockpit design. Height, chest, sleeve length, weight, leg length and so on were measured in the study.

What Daniels found in his study was striking. Not one single airman out of the 4,063 airmen he studied was “average” across the ten dimensions. Crazy. He then wondered how many airmen were “average” across just 3 of the 10 dimensions? Only 3.5% of the pilots.

So what happened when the Army designed the cockpit for the “average” airman? They designed it to fit no one. The answer was to stop building for the average and make the pedals, seat, and straps adjustable to the individual. Problem solved.

Last week, we talked about a guy who’d never farmed, who won a yield contest with AI, and I promised you the single biggest mistake people make with AI, and how to fix it. Here it is.

You open ChatGPT, or Claude, or whatever you’re using, and you type something like:

How much nitrogen should I put on corn?

Eight words. You hit enter. You get back a paragraph that sounds smart, throws around some numbers like 180 to 220 pounds per acre, mentions yield goals and soil type in general terms, and reads like a page out of a college textbook. It’s not a bad answer, but what if the corn was following soybeans and you put out two tons of chicken litter in the spring? The answer would most certainly be different. Essentially, what you’ve done with that prompt, is build the cockpit for the “average” airman.

The thing nobody tells you about LLMs

These tools don’t know anything about your farm or business by default. Not your soil. Not your rotation. Not what fell out of the sky last weekend. It doesn’t know the CEC of your soil. It doesn’t know what you planted.

When you ask an eight-word question, the AI does exactly what it was designed to do. It gives you an average of corn nitrogen recommendations from Florida to Minnesota. That’s literally what’s happening under the hood. It’s pulling the middle of a bell curve that contains an Iowa farmer with black dirt and 250 bushel yields, a dry-land farmer in Alabama, a research plot in Indiana, and your cousin in Tunica county all mashed together.

But, the middle of that bell curve is worthless to you, because you don’t farm the middle. You farm a specific field, with specific dirt, and a specific situation, and the only answer that helps you is the one that comes from your corner of the curve.

The 90-second habit

Before you ask the AI a single question, spend 90 seconds telling it what it would need to know if it were standing in the field with you. The nerds call this context engineering.

That’s it. That’s the whole habit. I’m calling it the 90-second habit on purpose. Any longer than that and you probably won’t do it. Any shorter, and there’s not enough information for the AI to actually help you.

Ninety seconds is the sweet spot where the friction is low enough to build the habit. You can do it standing on a turn row. You can do it at the kitchen table while the coffee’s brewing. You can do it from the cab.

What goes in those 90 seconds depends on the question. For a nitrogen call, it’s soil type, what nutrients you’ve already applied, growth stage, recent rain events, yield goal, past crops, and anything else specific to your farm and your situation. For an insurance question, it’s your APH, your county, your unit structure. For a marketing decision, it’s your basis, your storage situation, how much you’ve already priced, how far you are from the elevators. The category changes. The habit doesn’t.

The pattern is always the same: tell it what you know, then ask what you don’t.

But gosh, Mark, that gets old and time consuming

It absolutely does, and that’s exactly why we are going to make some context permanent. In the next few steps, I am going to show you exactly how to make some context permanent in your claude or ChatGPT account.

The Set Up

I was curious, and asked claude to estimate how many people actually complete this step, and the estimate Claude gave was around 10%. I was shocked. You will immediately see better results after you complete your instructions for Claude and ChatGPT. So, let’s get to it.

1. First, I’m going to walk you through how to set up instructions in Claude. If you use ChatGPT, skip to section two. First, click the three small lines in the top left corner of your screen. Then, click the small circle with your initials in the top right corner. Click on Profile. Then click into the personal preferences section.

2. Open your ChatGPT mobile app (iPhone), click the two lines in the top left corner of your phone. After you do that, you will click the circle in the top right corner of the screen (I have a picture of myself). Click Personalization. Then click “Custom instructions.”

3. If that section is blank, we are about to level you up. I suggest that you add five sections to your Personal Preferences/Instructions: 1) Who I am 2) Current Projects 3) Goals 4) How to Respond and 5) Style. This is the way that I have set up my account and I think this is a very good baseline.

   1. The first three sections are self explanatory but I’ll give a little background so you will know where to start. I told it where I went to college, my degree, how long I’ve been farming, how I plan to use Claude, what I’m currently working and dreaming on, and my specific goals. When writing the instructions/ preferences, I use all cap headers for the paragraphs. WHO I AM, CURRENT PROJECTS, GOALS, HOW TO RESPOND, and STYLE. I do this so that the AI understands how I am organizing my instructions.

   2. For the last two sections here is my advice:

      1. HOW TO RESPOND- (this is a copy and paste from my actual preferences)

         Default to deep, technical answers at maximum signal density: compressed, no filler, as long as the substance requires and no longer. When depth and brevity conflict, depth wins. Be direct. When facts and evidence counter my claims or questions, confront me. I prefer high ROI opportunities. When I’m making a decision, give a reasoned recommendation with the logic fully exposed and the key uncertainties listed explicitly. Be honest about uncertainty in every answer. Never overstate confidence. Never invent a source or citation. When something has no citable source or sits beyond what you can verify, reason transparently from first principles and label it clearly as your reasoning rather than sourced fact.

         Generic, inaccurate, or unsourced arguments are useless to me.

      2. STYLE- Don’t overuse the “-” character. I’m the audience unless I specify another audience in a project. I’m a farmer. Make sure that your answers are relevant to my farming context.

   3. You should absolutely tailor these sections to you and your preferences. I change my personal preferences at least once a week. The important aspect of preferences is to make it tailored to the information you want. You create the customized cockpit, make sure the information you are giving will provide context for the answers you want.

Pulling it all together

Gilbert Daniels found that the Army had built its cockpit for an average pilot who did not exist. Out of 4,063 pilots, not one was average across all ten dimensions that mattered. So they stopped fitting the man to the cockpit and built a cockpit that fit the man. Crashes dropped.

Your default AI is that fixed cockpit, built for the average. Setting up your AI with your context and implementing the 90 second habit is how you make it adjustable, so the answer fits your operation instead of a curve you don't farm.

This is your farm, your business, your dirt, your situation. Don’t settle for answers from the middle of the curve.

Next week: we go deeper in providing the right context for your work. It’s a game changer. Can’t wait.

Here’s an article I found fascinating:

Derek Thompson

The 26 Most Important Ideas For 2026

Thank you all so much for reading and subscribing to this newsletter. It’s been an absolute blast whose success in its first six months has tremendously exceeded my expectations. If you’re arriving for the first time and like what you see, please subscribe…

Read more

9 months ago · 2497 likes · 147 comments · Derek Thompson

Share

This is a conversation space exclusively for subscribers. Think of it like a group chat or live hangout. I’ll post questions and updates that come my way, and you can jump into the discussion.

Join chat

How to get started

1. Get the Substack app by clicking this link or the button below. New chat threads won’t be sent sent via email, so turn on push notifications so you don’t miss conversation as it happens. You can also access chat on the web.

Get app

2. Open the app and tap the Chat icon. It looks like two bubbles in the bottom bar, and you’ll see a row for my chat inside.

3. That’s it! Jump into my thread to say hi, and if you have any issues, check out Substack’s FAQ.

When it was over, Nipuna Chamara, the research assistant who led the team that won the trial, said something that gets right to the point:

I farm cotton, corn, soybeans, and wheat on roughly 3,000 acres near Friars Point, Mississippi. I’ve been using tools like ChatGPT on real decisions for a couple of years now. Variety selection, reading soil reports, sorting out crop insurance, checking chemical labels, and more. My goal with this newsletter is to show you exactly how I do it, step by step, in plain language. You don’t need to be a computer person. If you can send a text message, you can do this.

First, let’s kill the hype

Chamara didn’t win because the computer is smarter than a farmer. He won because he figured out how to leverage a tool.

The AI had no idea what was happening on his plot. None. He had to hand-feed it everything. He entered the soil type, moisture, weather, growth stage, and more. The first year he tried, in 2024, he basically gave up because he wasn’t seeing good results. It took him another full season of learning how to prompt/ask before it worked. (When I say prompting, I just mean the question or instruction you type into the AI)

If you’ve never touched AI, download ChatGPT, Claude (Anthropic), Gemini (Google), or Grok (xAI) today. Or you can just go to chatgpt.com and explain exactly what you’re after. This whole series is built around helping you use it better on your farm, in your business, and at home.

And if you already live in ChatGPT and Claude all day, stick around. I’ll show you the advanced stuff most people never get to. Like how I load a massive amount of data into a Claude Project, chain prompts into repeatable workflows, build my own tools with Claude Code, and use AI agents.

So, what’s it actually good for, today?

I could give a thousand ways I’ve used it in the last month. But being specific, giving it as much context as you can, (I often get better answers in voice mode, just talking to it), and asking for its reasoning, is a great place to start.

Here’s the honest scorecard.

I believe AI is the most powerful tool we’ve had since the creation of the internet. We must first learn how to use it.

The one thing that separates the winners from the frustrated

Everybody who tries one of these tools and quits says the same thing: “it just gave me generic garbage.” Yeah, because you asked it a generic question.

The contest winner’s edge was never intelligence. It was context. He told the machine everything about the plot before he asked it anything. The farmers and business leaders who get value out of AI are the ones who can say here’s my soil, here’s my rotation, here’s what I sprayed and when, here’s when I planted, AND ONLY THEN, asking a sharp question. The ones who stay frustrated type “how much nitrogen should I put on corn” and expect a miracle. Same tool. The difference is what you bring to it, which, if you think about it, is exactly like everything else in farming and business.

Why do we even need this tool?

We are all faced with decisions every single day. Most of the time, we do the best we can with the information that we have available to us. If you are anything like me, a tad bit scatterbrained, recalling information from last week can be quite the task-especially if we are talking about last year or a decade ago. I blame the concussions. But when used correctly, AI can hold massive amounts of data simultaneously and act as filters for our questions.

The truth is, farming has never been short on data. For the past decade we’ve been trying to figure out what data we need to be tracking -yield maps, fertilizer techniques, irrigation practices, soil samples, tissue samples- we're drowning in data. AI doesn't replace the judgment we've earned in the field, it just makes sure the right facts are sitting in front of us when it's time to use that judgment.

But can it replace my agronomist?

Let’s not oversell it. Your agronomist walks your fields. He knows your dirt. He remembers last year’s spring rains and the lessons that came with them. He can stand in a cotton field full of thrips and tell you something no chatbot ever will, because the chatbot isn’t there.

AI doesn’t replace that man. It makes him sharper. With AI, he can filter all of his questions through the resources that he already uses (more on this in issue 3 and 4). The tool is built to make us more productive, and more productive we will be if we learn to use this tool well.

The future isn’t AI or the agronomist. It’s actually AI, the agronomist, and the farmer, all using a new tool that leverages our collective knowledge. That’s the thesis of this whole series. Over the next few weeks I’ll show you exactly how I use these tools on my own operation. I’ll show you the prompts I use, the workflows, where they break, and what they’re worth in real dollars. No theory. No hype. Just what I’ve seen working on 3,000 acres of Delta dirt.

Next week: the single biggest mistake people make with these tools — and the 90-second habit that fixes it.

Sources & further reading

• UNL — Here’s the article about Chamara and the corn trial

• WSJ — On AI’s potential —AI solves a decades-old Erdős math problem (requires WSJ subscription)

• The AI Daily Brief (This is a podcast that I listen to often