“Scientific progress is one essential key to our security as a nation, to our better health, to more jobs, to a higher standard of living, and to our cultural progress.” -Vannevar Bush, 1945

Science is the foundation of progress, the mother of abundance, and the father of possibility. Like magic, science allows us to manipulate the world in ways we didn’t think possible, whether it’s talking to someone on other side of the globe, soaring into the sky like birds, or taking a pill that cures a disease.

When I was in college in the late 90s, I expected that by the time I hit 40, we’d have a cure for cancer, we’d have unlimited clean energy, universal high-speed internet, and a lunar base. Sadly, none of that occured. Despite major achievements in areas like computation, the internet, and mobile over the past few decades, we shouldn’t become complacent about the march of progress.

Thus, I sought to understand why we don’t see more science commercialized. Sadly, the research on how science is commercialized is rather sparse. The Institute for Progress is doing some great research in metascience, headed up by Heidi Williams. J Storrs Hall wrote a fantastic book Where’s My Flying Car. Eric Gilliam has a whole substack dedicated to the history of innovation. But overall, questions like “how well is technology being spun out of National Labs?” often go unanswered.

My main thesis going into my research was that we aren’t commercializing enough science out of research institutions like universities and national labs. All evidence I’ve seen has corroborated that thesis, though I’ve also uncovered many other factors: are we funding the right research? are we coordinating properly across institutions for moonshots? is government funding crowding out other sources of capital? are VCs the wrong type of capital to fund science? why aren’t entrepreneurs seeking out technology from labs to turn into products? is society ready for new technology?

What’s gone wrong? I’ve spent the better part of a year researching these questions and aim to share my findings in this post in hopes that I’ll find kindred spirits.

The Conditions for Progress

If we zoom out, I posit there are a number of societal factors that affect scientific progress:

1. R&D — we need to invest in foundational science
2. Commercialization — we need to translate science into products
3. Culture — people need to be excited about technology change
4. Regulation — we need a permissionless regulatory environment for emerging technology

Culture and Regulation are deeply interlinked. With a permissionless regulatory environment like the explicit laissez-faire policy of the Clinton administration in the early days of the Internet, we created enormous value for humanity. In the following decades, we democratized information, got mobile phones into the hands of more people than have clean drinking water, and created artificial intelligences. Compare that to the Nuclear Regulatory Commission (NRC), which effectively halted progress on nuclear power. Instead of entering the 20th century with a network of advanced, zero-carbon nuclear reactors, America still belches out carbon from coal plants, a technology that was remarkable in the late 19th century, but continued long after its ideal lifespan.

Though permissionless innovation is important, the effect of culture cannot be discounted. The NRC was timid about nuclear power exactly because the broader population was terrified (rightly) of nuclear war and fearful (unfortunately) of nuclear energy. Movies like The Day After (1983) and The China Syndrome (1979) were cultural lighthouses, guiding us away from the positive aspects of nuclear energy. The China Syndrome premiered just a few weeks before the Three Mile Island nuclear accident, helping to fuel ticket sales and stoke the flames of fear. The dawn of the AI age is looking much more like nuclear than the internet, with society wringing its hands in fear of AI overlords, instead of imaging the incredible possibilities for unlimited intelligence.

Likely there is work to be done on understanding the relationship between culture and innovation. Culture is the tail that wags the dog of government: if people are generally positive towards technology, then government will follow suit and play its proper role of maintaining rule of law and amending regulations to support new products. Marc Andreessen’s Techno-Optimism Manifesto spoke to people like me who see positive upside in technology, but failed to recruit new people to the cause.

Cultural acceptance of technological change an investigation for another day and for another person, as I’m hardly someone who understands cultural trends.

R&D and Commercialization are also a feedback loop. If we spend huge amounts of money on research, which don’t translate into products, then we will further our intellectual understanding of the world but won’t get the fabulous benefits of that understanding. On the flip side, science that is connected to real world problems will direct our brightest minds towards science that can radically improve society and industry will work quickly to translate the science into utility.

So what’s going on with that feedback loop? My intuition is that too much technology sits on the shelves of labs instead of benefitting society. America has the largest R&D budget of any nation with 28% of the global total. Are we seeing society progress as fast as we might have expected?

How is R&D commercialized?

My experience is in building products from the raw materials of technology. Almost every startup I’ve been involved with has spun out of a research institution (Stanford, Xerox PARC, Los Alamos National Laboratory, UIUC) and I’ve seen the challenges of spinning something out, finding aligned capital who understands technology as well as business risk, and building a product out of raw scientific research.

Much of what follows is based on my own experience, but also has been corroborated by those that I’ve spoken with in academic, tech transfer organizations, and in deep tech investment.

The regulatory environment is a burden, particularly in highly regulated markets like energy and healthcare. But for the COVID-19 pandemic, we might have let the mRNA technology underlying vaccines sit in a quagmire of regulation. It shouldn’t have taken a global crisis to unlock the value of science.

Even if we were able to radically improve regulation, we’d still have a massive issue in jailbreaking technologies from research institutions.

The government is deeply embedded in R&D and this might have become a problem. The chart below shows the relative spend for basic research from the four key institutions of society, with government and education spending the bulk of dollars. Not only has the absolute number of government increased considerably, the relative share of the government now dominates. (As an aside, getting accurate numbers on science funding is extremely difficult: what counts as “basic” vs. “applied” research, for example?)

When government plays its proper role in science funding and experimentation, it can move the world forward. World War II was won by America’s ability to mobilize our production capacity. The US produced more tanks than all of the Axis powers combined and twice the production of artillery and machine guns. However, it was also won by technology: radar, sonar, missile trajectories, and, of course, The Manhattan Project. Seeing the success of science, the government took a much larger role in science funding. The Apollo Program was a response to the threat of the USSR dominating space and ushered in the space era. Dollars deployed from ARPA went into many different technologies, the most notable being the Internet (originally ARPANet). Organizations like Stanford Research Institute (SRI), which received huge amounts of government funding, created technology necessary for the PC, like the mouse and early graphical user interfaces.

Why was the government so successful during this period in deploying R&D dollars efficiently to great effect and integrating that research into society? One potential explanation is that the vision was very clear in all of these projects. In WW2, there was a clear and present danger of the Axis powers dominating the world with a totalitarian Nazi regime. After winning the war, we needed to deal with the threat of communism. The practical focus of needing to ensure that communication was maintained even in the face of a nuclear strike is what gave us the Internet. The reality of needing to reduce the weight on the Apollo spacecraft is what drove the desire to use new-fangled integrated circuits as computers, providing an early capital infusion into the chip market.

What is the current challenge the world is facing? China is certainly a threat, but there is no consensus on exactly what that threat is or how to deal with it. Climate change technology has received trillions of dollars of funding and all we’ve gotten is more efficient solar and wind, hardly technologies that unleash clean energy abundance of clean energy and (arguably) a poor return on capital given alternatives like nuclear. The Covid pandemic could have been a rallying moment in society to radically improve medical technology. Instead, mixed signals from the NIH has divided our country further, making Covid opinions political instead of scientific. Lacking a clear vision will hamper our deployment of R&D capital.

However, that lack of vision still doesn’t explain why given billions of research dollars every year, we don’t see more science entering society. The government realized this gap and passed the Bayh-Dole act of 1980, creating a structure to allow technology to be spun out of labs. The Act allowed institutions to profit off their inventions, even though the grant money was from the government. This was well-intentioned and based on capitalist principles.

Unfortunately, that act was also the progenitor of our current “tech transfer” infrastructure.

Technology Transfer Offices (TTOs) exist at most research institutions with the intention of licensing IP developed at the institution for the outside world. My personal experience working with TTOs is horrendous: they are typically lawyers who see their role as protecting IP rather than shepherding IP to its highest-and-best use. There are some bright spots in tech transfer, but the vast majority of TTOs should be shut down and rethought completely.

A key problem with research institutions is that the protected IP doesn’t have a clear and obvious use case. There are exceptions, like spinning out drugs. Those are easy, since once efficacy is proven, there are large drug companies with plug-and-play infrastructure to manufacture, distribute, and market the new drugs (not to mention deal with the regulatory burden of the USDA). Most technologies don’t have as clear cut cases. A personal example was Descartes Labs, where we spun out mathematical optimization algorithms for machine learning over complex signal data. Choosing satellite data was driven by the markets: since so many satellites were going up, our technology was well suited to analyze the enormous amounts of data coming off the hardware. That connection between the underlying technology couldn’t have been found by the TTO. Our team needed to go into the wild to find product-market fit.

Further, much of the technology at research institutions isn’t ready for venture capital. VC typically likes to invest in businesses with clear markets, clear products, that fit into established models, like software-as-a-service. VCs famously say that they don’t fund “science projects,” which is the majority of the IP produced by a research institution. At Bell Labs, for every Bill Shockley (transistor) or Claude Shannon (information theory), there were 9 research scientists, who were responsible for translating the technology into something that could be licensed externally or deployed within the AT&T network.

VC capital is adverse to deep tech, ince the LPs for VC funds demand certain kinds of returns. So, even though VC is technically risk-capital, most VCs act like PE firms, looking to reduce the risk of their investments through financial analysis. Since SaaS metrics and software financials are well understood, most VC capital goes to software companies. Atoms are forgotten when bits are easier to invest in. I posit that we need a new capital structure for commercializing science, probably patient capital from sources like university endowments, foundations, and permanent funds.

Though there are many problems with research institutions, business also doesn’t seem to be playing its proper role. The golden era of corporate research labs was in the middle of the 20th century: AT&T’s Bell Labs, IBM Research, Xerox’s PARC, Sarnoff, etc. Bell Labs alone is responsible for some of the most important inventions of the 20th century: the laser, information theory, the transistor, the underwater cable, satellite, cellular technology, to name a few. Not only were the inventions practical, but basic research needed to be done to enable these new technologies: Bell Labs received 9 Nobel Prizes. What can we say about the R&D budgets of modern corporate giants like Apple, Google, and Microsoft? Certainly DeepMind at Google has written seminal papers in AI, a big win. But, the most prominent and well-funded of research labs, Microsoft Research (MSR), has produced reams of papers but few notable innovations beyond incremental improvements to Microsoft’s software products.

I don’t have a strong thesis yet on why corporate research seems to have stagnated from the astronomical gains of the mid 20th century. Perhaps there is a vision problem again here. At Bell Labs, the goal of the organization was to improve “the network” and the AT&T end-customers were always in mind. Even theoretical advances were important to this vision, like when Claude Shannon created the mathematics of information theory. Not only was this a huge leap forward in quantifying information (e.g., signal vs. noise), it was immediately practical in devising ways for AT&T to compress and unpack information sent over their network. There was no such thing as theory divorced from practicality at Bell Labs.

Another possibility corporations focus their R&D budgets on near-term practical technologies since government R&D funding typically is focused on longer-term bets in science. A consequence of this is that there’s a missing feedback loop between practical applications and ivory tower science. Scientists don’t have to deal with the messy world of engineering their science into something practical, and certainly not worry about whether or not consumers or society needs what they’re building. In the 20th century, some of our brightest minds in physics gave us practical advances in materials and information transmission. Now, they work on abstract concepts like string theory and the Standard Model in physics. When we split the atom, there was immediate utility. When we discovered the Higgs Boson, we patted ourselves on the head because the mathematics worked out, but there was no clear utility.

Solutions?

Abundance comes from translating science into practice through engineering. Despite the US (and the broader world) spending huge sums of cash on science, we don’t see enough breakthroughs, nor do we see the breakthroughs translating into solutions at a fast enough clip. Where should we look for a solution?

I hope that my musing above convinced you that there is something broken in the system itself. A first step might be researching the history of scientific research as an institution over the past 150 years and building a model of how it works today: who are the players? where does the capital come from? what’s the process moving science from labs into society?

With that current state of the world well-described, we then could create a vision for a future state where scientific advances are regularly advancing society. This will require new research organizations, new capital structures, and a renewed role for business in scientific research.

My personal passion project over the past few years has been noodling an idea for a science accelerator called Morton Labs, to bridge the gap between research institutions and capital markets. If Universities produce IP that isn’t yet fundable, we need to solve the research engineering challenge (how to derisk the technology) and the market engineering challenge (how to create sustainable businesses with products that address a market need). Such an organization would require relationships with multiple sources of IP and a pool of extremely patient and risk-tolerant capital.

However, there are lots of other ways to think about solving the problem. Organizations like the Arc Institute, funded by Patrick Collison (Stripe) and Vitalik Buterin (Ethereum) are attempting to give grants to bold scientists over 6 years to come up with novel solutions in biology and medicine. Ben Reinhardt’s Speculative Technologies attempts to replicate the successes of the DARPA model — particularly its ability to coordinate resources across multiple institutions — to fund long-term speculative research. And others, like Eric Gilliam, are dreaming up new kinds of deep tech VCs.

This post wasn’t meant to have all the answers. In fact, it’s really meant to be a beacon for others who see the problem similarly. Whether you’re a university president, a program manager at NIH or NSF, a lab researcher, a disgruntled TTO employee, or a scientist hungry to change the world… I want to talk to you.

Society has built up strong infrastructure around science over the past century: we have the labs, the scientists, and the money. If we can build a vision for how science can positively affect society, then by the time we hit 2050, the world will be radically different and radically better.

Being CTO of a large community of nonprofits means that my team can explore the frontiers of technology. I’m excited to announce that Stand Together will be launching two new experiments in decentralized technology, known as web3, alongside our partners, The VELA Education Fund and Sky’s The Limit. We believe that emerging web3 technologies have the potential to encapsulate our principles into software, making philanthropy a more bottom-up process.

Although many people see the wild side of web3 as NFTs and cryptocurrencies, my team at Stand Together has focused on web3’s potential to coordinate large groups and distribute capital more efficiently than ever before. In this post, we’ll describe why we think community grant-making is a perfect web3 use case, why we think building in web3 drives innovation in philanthropy, and show how these experiments put our principles into action.

Community knowledge in grant-making

Grant-making nonprofits do just that: they give out grants.

The organizations raise money from donors, accept applications from potential grantees, and then go through an internal, top-down process to decide whether to give money to the grantee. A well-run grant-maker will track how well its grantees perform and use those data points to improve their selection criteria and process, thus amassing a certain kind of domain knowledge.

Here at Stand Together, one of our guiding principles is that we prefer bottom-up solutions rather than top-down solutions. Under what conditions could grant-making become a more inclusive, democratic, and bottom-up process?

Imagine that, instead of a centralized organization making decisions, we outsourced that decision-making to the community; and delivered superior results.

To make this more concrete with examples, let’s introduce our partners.

VELA Education Fund was launched in 2020 to accelerate education innovation. They’ve given out thousands of micro-grants from $2,500 to $10,000 to educational entrepreneurs, whether that’s students, families, educators or community members. VELA was inspired by the Fast Grants model pioneered by Tyler Cowen and Patrick Collison with the theory that the grant application process should be simple and straightforward and that it’s better to support many innovations and back the ones with the most promise.

Right now, VELA is distributing millions of dollars every year with a tiny team. Not only could they scale up more quickly by letting the community decide on grants, they hope that the community will strengthen the proposals by integrating local knowledge, leading to more impactful investments. After all, those that are closet to the problem are the ones with the best solutions. (If you want to geek out, I strongly recommend reading Hayek’s classic paper, “The Use of Knowledge in Society,” whose knowledge problem was inspirational in this project.)

Imagine that there’s a proposal to improve STEM education through outdoor activities. After that proposal is submitted, other STEM education experts in the community weigh in and the proposal morphs into something stronger based on that feedback. After that feedback is integrated, the community votes and decides to fund the proposal. Perhaps even those who had lent their expertise then follow the project and continue to support even after the grant has been made, creating peer-to-peer connections and increasing the likelihood of success.

We still have to work out the details, but that’s exactly the kind of system we want to build for VELA. Someday, VELA hopes that all grant-making decisions are made by the community.

Our second partner is Sky’s the Limit, a nonprofit that empowers underrepresented entrepreneurs, primarily by connecting them with business mentors. They also help entrepreneurs access startup funding by giving out a $2,500 grant every month from a “Friends & Family” fund, which has distributed over $300k in startup grants. To decide who receives grants, Sky’s the Limit uses mechanisms in their web2 community to accumulate points, and those points can be used to vote on entrepreneurs’ pitch proposals. The proposal with the most votes wins.

Sky’s the Limit believes that when people feel more ownership over their communities, they are more likely to take care of the community. Their innovative point system aligns incentives so that the members who create the most value for others in the community have the most say in who wins the grants. In Sky’s the Limit’s case, they want to increase the size of their grant fund, give out many more grants, and create a real sense of community ownership, while incorporating the expertise of their entrepreneur and mentor community.

Fun fact: in 2018, Sky’s the Limit designed, built, and launched a private blockchain on IBM’s Hyperledger fabric in partnership with and funded by Accenture’s Blockchain Labs. That experiment was years ahead of its time.

We all believe that the time is now for web3, so we’re going to build some experiments to bring community knowledge into the grant-making process.

Why build in web3?

You don’t need to be a web3 expert to understand the problem we’re trying to solve, but it’s important to understand the basic concepts of web3 to appreciate why we’re so excited to experiment at the bleeding edge of technology.

At its core, web3 is a collection of technologies that allow for decentralized coordination. You’ve probably heard of — and maybe even own — cryptocurrencies like Bitcoin. What makes Bitcoin so remarkable is the system to verifies who you are, how much money you have, who you are sending it to, and executes the transaction — all without a central authority blessing the process. Bitcoin is truly digital cash. Contrast this with making a payment with your Visa card. You have a physical card that you tap onto a machine, which sends a signal to Visa’s centralized system that authorizes you to spend that money with that merchant.

Core to making cryptocurrency work is the ability to establish trust digitally. The mechanics honestly don’t really matter unless you’re an entrepreneur or developer. What’s nifty about web3 is that it decentralizes coordination in many domains , not just money. This will challenge our notions about how humans exchange value, assign value, and organize ourselves. If that sounds abstract, it’s because web3 is a general-purpose technology with many different specific applications. As such, web3 has the potential to be really big.

In the case of decentralized grant-making, we are going to use a web3 theme about reshaping organizations: the Decentralized Autonomous Organization, or DAO. In a DAO, all decisions, including how a DAO is governed, are made by the community, making them inherently bottom-up.

A key use case for DAOs has been distributing capital. The Constitution DAO raised $41 million to buy (though unsuccessfully) a copy of the US Constitution and Ukraine DAO raised money to help the people in Ukraine. We were particularly inspired by Big Green DAO, a web3 experiment connected to Kimbal Musk’s Big Green, focused on local food sustainability. Big Green DAO allows the community to make grant decisions and has been evolving its governance over the past year. These early initiatives have shown us how web3 tools have made it easy for thousands of people collectively to raise and distribute millions of dollars.

The reason we decided on using a DAO rather than building in more traditional “web2” technology is:

• DAOs are focused on distributed coordination — DAOs are experimenting with how to coordinate a large number of people on a single problem. We face this very problem every day, and solving it is critical to our success.
• DAOs have a standard software stack corresponding to our needs — many DAOs raise a treasury, accept proposals, debate proposals, vote on proposals, and distribute funds. These are the functions our partners need.
• The DAO community is innovating on governance — DAOs are trying to coordinate hundreds and sometimes thousands of participants, they are innovating on all kinds of distributed governance structures. We will benefit from what the DAO community has learned already. Plus, we believe that the principles of human progress and liberal democracy will help inform DAOs innovations.

Through our experiments, we hope to demonstrate a real use case for web3 technology in philanthropy.

If you’re still not convinced on web3, don’t worry. Our biggest opportunity in these projects is to create an experience layer that makes it easy for everyone to interact with — and benefit from — web3.

What’s next?

Like any experiment, we are excited about the potential but humble about what we still don’t know. Web3 truly is a new frontier and we have a suspicion that philanthropy could be a major web3 use case, but that’s a thesis that needs to be proven.

Through our partnerships with Gotham Labs and Bankless Consulting, we have engaged experts within the web3 ecosystem.

If you’re a member of the web3 community and would like to support these projects either with your brainpower or by donating to the treasuries, we’d love to hear from you. And, if you’re a nonprofit who is considering a web3 experiment of your own, please reach out to us so we can compare notes.

Lastly, a key part of the web3 ethos is transparency. Expect us to report out regularly on our progress with both projects.

Oh, maybe most importantly, my team is hiring. We’re currently looking for a web3 product manager and a web3 futurist, in addition to a few other roles on my team (UX designer, Director of Product Management, & Senior Product Manager). If you’re a technologist who wants to apply your skills to the social sector, reach out for a chat. My DMs are open.

Before a board meeting early in the history of Descartes Labs, I sat looking at my deck, realizing that it lacked a narrative. I sat down and penned my first board memo and produced one every quarter thereafter. I named it “Meditations” after Descartes’ most famous book, in which he secured reason in the eyes of God, providing the philosophical foundation for the scientific revolution.

Today calls for a different kind of Meditation, a song of lamentation.

Descartes Labs was on a mission to better understand our planet through satellite imagery. To enable such a lofty mission, we built a data refinery, a petabyte scale repository of satellite and other geospatial data combined with petaflop scale supercomputing power. Because of our platform, we attracted some of the best scientists from around the world, often fleeing universities and government institutions, looking for an intellectual island to do impactful science. Our Cartesians built some of the most remarkable demonstrations of technology I’ve ever seen in my career.

Joe Morrison on Twitter: "I hope Mapbox and Descartes Labs alumni will be like the PayPal mafia for the geospatial industry over the next couple of decades. So many ABSURDLY talented people who created some truly special products. Those folks can build just about anything they put their mind to. / Twitter"

I hope Mapbox and Descartes Labs alumni will be like the PayPal mafia for the geospatial industry over the next couple of decades. So many ABSURDLY talented people who created some truly special products. Those folks can build just about anything they put their mind to.

Even with almost $100m of invested capital (strangely, Crunchbase doesn’t list the last two rounds), >$200m in total going into the company (revenue+investment), ending 2021 with over $17m in revenue, and multiple 8-figure government contracts, the company was sold in a fire sale.

Writing that sentence is absolutely shocking to me. How is it possible that a company with an incredible team, so many successes, so much revenue, and so much invested into the underlying technology could possibly be worth close to nothing?

Given the wild ride of Covid and the economic rollercoaster ride we’re currently on, it’s easy to dismiss Descartes Labs as a victim of macroeconomic circumstances. I don’t believe that to be the case and will make the argument that there were two main reasons for the mismatch in the actual value of the company versus the price that was paid:

1. The company was burning too much cash.
2. The sales process was run poorly. (i.e., the process of selling the company)

At fault is the management team, who executed poorly, and especially the board, who knew these facts and chose to do nothing.

I’ll first dive into a bit of history about the company, since for the first five years — through a series of successes and failures — I was at the helm. After I was fired in January 2020, I have much less visibility into the day-to-day of the company, but will do my best to assess what happened.

Of course, I’m telling the story from my vantage point. I’ll do my best to be objective and honest, but any story written by a passionate founder ought to be read with a skeptical lens. I do hope we hear other perspectives and perhaps an objective third party can take on the challenge of synthesizing into an approximation of the truth.

I’ve been warned about writing this post, “You’ll never raise money from the Valley again…” The power dynamic between founders and VCs is asymmetrical. VCs have no problem firing a founder and being vocal about it, but when founders observe bad behavior from VCs, the advice is to shut up and find better investors next time.

The SAFE Note changed seed funding by preventing bad early-stage investors from injecting punitive terms into convertible notes. I believe there ought to be something similar for founders: a set of standards around compensation, severance, stock, and board structure. I’m going to make it my mission to collaborate with other founders and the VC community to build some kind of Bill of Rights for Founders.

Until then, here’s the story of Descartes Labs, how a remarkable geospatial startup went from flying high in orbit to end in a fiery crash back to Earth.

Anni Mirabiles: The Early History of Descartes Labs (2014–2016)

Descartes Labs defied common wisdom in our early days. We spun out of a National Laboratory, we focused on science instead of engineering, we chose to headquarter in New Mexico before it was popular to flee California, and we were able to attract coastal money to a state that had never produced a successful software startup.

After I helped sell Zite from CNN to Flipboard in March of 2014, I had a fantasy about going on sabbatical. After some travel, my plan was to move to NYC to do part-time consulting while I figured out my next startup. As fate would have it, I got a call from a friend who said that one of his friends (a semi-retired VC living in New Mexico) had been working with a group of scientists at Los Alamos National Labs (LANL) who were building a search engine for the video archives of large media companies. Imagine typing in “iraq bomb explosion” or “George W. Bush blooper” and getting hundreds (or, in the latter case, thousands) of results. Given my background in search, my 3 years at the helm of Zite within CNN, and my penchant for spinouts, it seemed like a perfect fit.

The market wasn’t responding well to their pitch. The scientists had been trying to raise money from VCs with fits and starts for at least a year before I met them. There was a term sheet from a prominent VC for a $15m series A, but those conversations and many others fizzled out.

Though I, too, was extremely skeptical of their business plan, I hopped on a plane to New Mexico on Bastille Day 2014 because I had been obsessed with The Manhattan Project since I was a little kid. When I met the cofounders, I was blown away by their brilliance and humility. Immediately after that first meeting, I knew we needed to start a company.

By the end of 2014, we had settled on satellite imagery instead of media search, we created a company, gave it a “temporary” name of Descartes Labs, raised $275k in angel capital by October, signed a deal with LANL, and closed a $3m seed funding with Crosslink Capital in December of 2014.

When VC is at its best, it deploys its capital against world changing bets, even in the face of uncertainty. Our deck basically said “We have some science! We have some smart people! Satellites are going to generate a lot of data that we think will be useful! Here are some verticals who might care!” The fact that we were able to circle over $3m of capital on such a risky proto-company gave me hope for a better future through technology.

Armed with a small warchest, I hopped in my car with a friend and drove to New Mexico to begin a new adventure.

For the first few months of 2015, the first order of business was setting up our digital laboratory. We got $100k in startup credits from Google Cloud Platform (GCP) and started collecting satellite data to run experiments. In March, my cofounders came up with what I thought was a wacky idea: they wanted to process a petabyte of data in under 24 hours, which would burn at least a quarter of our credits. For context, a petabyte is a million megabytes; an average photo on your phone is around 2.5MB. To me, it just seemed like a distraction, since there was no clear customer value out the other end; on the other hand, when you’re building a Lab, sometimes you just need to let the scientists play.

In mid-April, despite all odds, our team commanded 30,000 processors for 16 hours to complete what became known internally as “The Big Run.” I was absolutely astounded. So was Google. We got a call early on in the run because they thought something was very, very wrong, like a denial-of-service attack. After we begged Google not to shut down our calculation, we became friends. They sent out some folks later that summer to chat with us, beginning a long and fruitful partnership with their team, culminating in a visit by legend Vint Cerf in 2019. Not only were we the first startup in history to burn through our startup credits, we became the Southwest’s largest GCP customer. Of course, this isn’t necessarily a good thing because, as you’ll see, our cash burn contributed to our demise.

(As a side note, Descartes Labs has been on the Top500 list of the world’s fastest supercomputers since 2019, most recently at #40, thanks to the monumental work of a co-founder.)

Now that we had done great science, it was time to take that computing power and data and build customer value. Hedge funds were a target in the early days of a company and we learned that they were hungry for alternative data sources to give them an edge, something still unique in 2015. Predicting global agricultural yields seemed like a reasonable use case given our core dataset, so we began Project Iowa, an attempt to build a corn-yield forecast for the US.

And did we ever! After releasing our predictions ahead of a very important US Department of Agriculture (USDA) forecast, we singlehandedly moved the market 3%, thanks to a well-timed Bloomberg article. Even though we moved the price in the wrong direction (2015 wasn’t our most accurate prediction), it didn’t matter. Big agricultural companies who weren’t returning our phone calls started answering and we were able to raise another $5m from AgTech VC Cultivian-Sandbox.

In 2016, we focused on selling our global agricultural yield forecasts. Cargill, the largest agricultural company in the world and largest privately-held company in the US by revenue (~$130b), approached us with a challenge: could we combine our data with their proprietary data to create better models for their agricultural supply chain business. After many months of hard work, Cargill decided to partner with us, becoming both a customer and an investor.

It’s hard to understate the importance of Cargill to Descartes Labs. Not only did the relationship provide revenue to the company, allowing us to raise a Series B and hire a lot more people, but the thought partnership between Cargill and our team helped us to figure out who we were. They realized before we did that the magic of Descartes Labs wasn’t in a SaaS product to democratize satellite imagery. No, we were a modern AI consulting company (cf., Palantir) who had assembled brilliant minds, built an internal set of tools that gave our scientists access to huge amounts of data, and created a collaborative internal environment to solve complex problems. You can see me and James Weed of Cargill speaking at Fortune Brainstorm Reinvent in 2018.

Another important relationship that began in 2016 was DARPA. We were awarded a multi-year $1.5m contract to investigate how geospatial data might be used to forecast famine in the Middle East & North Africa. Not only were we helping Cargill to feed the world, we started working on important social impact work.

By the end of 2016, we were feeling pretty good about defying the odds of being a New Mexico spinout of a National Lab. We were even profitable in Q4 2016, unheard of for a science startup.

Descartes Labs had become the Miracle on the Mesa.

The Descartes Labs Dialectic: Palantir or Salesforce? (2017–2019)

All the while, a debate was brewing within the company, known as the Descartes Labs Dialectic: are we building a product or a services company?

Descartes Labs crossed $10m of revenue in 2017, typically a massive milestone for software startups. The problem was that 90% of our revenue was in just a handful of accounts. Thanks to a brilliant cofounder in sales and a technical team to back up whatever custom deal was sold, we scored some really big wins for ourselves and for our customers.

The problem was that we hadn’t built any products yet.

Our corn (and eventually soy) forecast was a product of sorts, but we had trouble selling it. Proving that it had “alpha” (i.e., an edge that gives a trader an information advantage) was difficult and the value of a straight forecast was limited. I came to believe that selling data products was the wrong business model for AI startups. (A post from a16z penned by two of their partners in early 2020 is an even better statement of the sentiment.)

Another potential product was the underlying data refinery. Perhaps we were building a platform? We used it to great effect with our customers in building custom models for them. But, the companies that could derive the most value out of our platform were typically in commodities like agriculture, shipping, metals & mining, and forestry. They didn’t have the technical expertise to extract the full value out of the data or computing power within our platform.

We, like many AI companies like Palantir, were a hybrid consulting company: we built a robust platform that we were the best in the world at using and charged our clients lots of money for building unique, extremely valuable, proprietary (read: cannot be sold to others) models. Even if we structured the revenue cleverly by selling our customers a platform subscription and subscriptions to the models we built for them, we still weren’t a SaaS company.

If I could go back and change just one thing, it would be the resolution of the Descartes Labs Dialectic. I would have shut down internal debate in the company from the crowd that thought we should be building a software company. I would have been much clearer in our fundraising decks about what strategy we were pursuing: AI companies can build an enormous amount of enterprise value though specialized consulting contracts.

I succumbed to the market narrative, pushing startup founders to pursue a Software-as-a-Service (SaaS) business model.

To “solve” the Dialectic and package Descartes Labs as a software company, we separated the team into the Platform Team and the Applied Science team. The Platform Team built the tools and the Applied Science team built the models. Our typical customer engagement started off with a pilot project, usually around $100k, and our theory was “land-and-expand” (another popular SaaS trope): grow the small account into a much larger account, paying >$1m / year. Even though we were taking consulting contracts in the short term, in the long term, a product would emerge from the engineering team.

Or, at least that’s the theory on which we raised our $30m series B from March Capital in the Summer of 2017. Series B was another pre-emptive round, raised without me having to go through a full process and shake the money trees on Sand Hill Road. This new capital was intended to grow our sales & marketing team, build out our “product,” and move into a growth stage so we could raise a healthy series C.

Looking back on Series B, I’m conflicted. On one hand, I absolutely believe in taking money when it’s available. In 2017, we were 8 years into a bull run and we wondered how long the gravy train would keep running. On the other hand, when you take a big round, it’s important to be prudent with the cash and expand only when you believe that there is a viable business model. This requires investors and the company to be aligned around what constitutes product-market fit and what signals indicate that the business model is ready to be scaled.

In 2018, we scored a few minor wins, which kept our revenue up and to the right, close to $20m. Thanks to superior performance against our original DARPA contract, we were able to translate that contract into a much larger deal to build a geospatial data refinery for the government, in a contract worth up to $7.2m. We also made a considerable amount of progress booking pilot contracts from $50-$150k, not bad for initial projects. For 2019, our plan was to continue that pipeline of pilots and translate some of those pilots into large, multi-year 7-figure contracts and we’d be flying high.

However, by the end of 2018, we realized that our land-and-expand thesis was encountering roadblocks. We were losing money on pilots (read: negative gross margins) and translating those pilots into larger deals was difficult. Predicting revenue was nearly impossible (read: not a repeatable business model), driven by long lead times and uncertainty around how much value our pilot projects would deliver to the customer.

Things got dicey around our Q4 2018 board meeting, where we presented our 2019 plan. We projected that 2019 would be a 50% increase in revenue year-over-year, given that we hadn’t quite figured out our sales process or product . The board was not pleased, they wanted to see a much steeper growth curve.

Now the Descartes Labs Dialectic reared its head. Our investors wanted us to be a SaaS company with SaaS metrics and SaaS growth. We simply were not. We should have structured our entire business around being a high-end consulting company. Perhaps we wouldn’t have gotten SaaS multiples. Perhaps. Or perhaps we would have focused our energy on what we did best. I’d rather have $100m / year of long-term consulting contracts than burning expensive venture capital on a fantasy SaaS product.

Ultimately we capitulated to the board’s desire for an unrealistic revenue expectation because, in my mind, isn’t that what I signed up for when I raised the money? Unfortunately, it caused the company to focus on improbable but high-value deals instead of getting our product philosophy and sales strategy in order. By pushing us into unreasonable growth expectations, the board drove us to hit our numbers in the short term, not build a long-term engine for growth.

2019 admittedly was not my best year as CEO. We had grown very quickly and our leadership at the time didn’t have enough experience to manage that growth. Further, hiring in New Mexico was extremely difficult. We got really lucky in hiring two experienced Silicon Valley executives who both came to New Mexico thinking they would retire, but were seduced by our mission. Each influenced me greatly and I’m thankful to this day for their wise counsel. That being said, I badly needed an operational CFO to help me understand the numbers. My comparative advantages are building a vision, imagining what products emerge from a set of technologies, and raising money for a company. How that money is spent is not my forte and I needed the right kind of experienced help. With a requirement that this individual had to live in Santa Fe, I missed out on a number of opportunities to build up my leadership team and didn’t find anyone until September 2019.

About halfway through 2019, the board and the company knew that hitting our sales numbers was unlikely, since the number was dependent on either closing some highly-speculative commercial deals and a single government contract. The commercial deals didn’t close and we lost the government contract, despite signals that we were going to win it continuing up until the 11th hour. Despite all of this, we were able to close another $20m investment in late 2019 to extend our runway at $200m valuation.

The board decided that it wanted to be “helpful” and deployed someone to be part-time embedded within the company. Despite limited operational experience, the “help” built a separate financial model that utilized SaaS metrics and expected us to abide by that model.

The board had chosen the wrong path in the Descartes Labs Dialectic. By golly, they valued us at SaaS multiples and we were going to be a SaaS company!

I got the call in November 2019 that the board had decided to fire me. The stated reasons were: our burn was too high (~$1m/month), we missed our sales projections for 2019, and I was slow to hire experienced executives. All of that was true. However, firing the founding CEO is a huge disruption to a company and needs to be thought out carefully. Not only does the founding CEO know the company better than anyone else in the world, but they have the incentive structure to fight like hell to make sure a company survives.

Many CEOs have issues when they cross 8-figures in revenue, balloon to over 100 employees, and are figuring out how to manage a company. To this day, I feel that the punishment simply doesn’t fit the crime and I’ve never gotten an explanation. The reality was that we have over $20m in revenue and when we needed to extend runway, I was able to raise more money. Descartes Labs was still, by any measure, an incredible success.

Their exit “plan” for me was rushed: they wanted me out within a few weeks, akin to walking me out the door. They were going to put the CFO, who had only been there two months, in charge of the company until they could find a new CEO, though they weren’t planning on telling the CFO that he was going to be temporary.

I was feeling burned out at the end of 2019 and would have accepted a transition plan where we agreed that I’d cut costs and aggressively search for a replacement CEO. That plan would have saved the company $10m over the course of 2020: I could have surgically removed both cloud and people costs from the company to reduce burn while we found new leadership.

Another option, completely rejected by the board, was to take a step back and decide whether we were even in the right business. Maybe we shouldn’t have gone down the SaaS route, reduced our headcount, and focused on high-margin consulting deals. Maybe the satellite data business wasn’t as robust as we had originally thought. Maybe our assumption about growing to $100m in ARR was wrong in the first place.

An option never considered was to sell the company right then and there. Given that we had just raised at a $200m valuation and that the market was still hot, I could have initiated an acquisition process and likely sold the company for greater than $200m, given our incredible team, client list, and demonstrations of scientific prowess.

The board resisted a dialog and I capitulated. Seeing that I had lost, I tried to play the game. I prepped an exit, hopped on a plane to Davos (we were a 2019 WEF Technology Pioneer) and passed the baton.

The Post-Founder Era (2020–2022)

My visibility into the company was limited for the next few years. Though I sat on the board, I did my best not to talk to current company employees, to give the new CEOs enough freedom to operate without a shadow CEO poking in from the outside. I joined the committee to select a new CEO, though I knew that I was only there ceremonially (make no mistake: the investors will choose your replacement) and to convince the incoming CEO that a large common shareholder wasn’t going to be a thorn in his side. We hired and installed a(nother) new CEO in July of 2020.

An irony of 2020 was that a major reason for firing me was having a high burn, slightly north of $1m/month when I left. Our burn continued uninterrupted in 2020. Indeed, the burn during the last quarter during which I was CEO was the lowest burn for the remainder of the life of the company.

A condition of the new CEO joining was that the existing investors would invest another $20m (somehow, neither this nor the $20m I raised at the end of 2019 are on Crunchbase), which made sense given the need for more runway, but also added significantly to the preference stack. Also, it tapped out our investors–they had no more money they were willing to commit to the company. For example, Descartes Labs was the single biggest cash investment in the history of Crosslink Capital and they couldn’t write another check.

I was asked to leave the board at the end of 2020, which worried me terribly since there would be no common shareholder or founder representation on the board. Since I had a chunk of stock options that would expire within 90-days without a relationship with the company, I brokered a deal to leave quietly in exchange for maintaining those options until the sale of the company. Plus, it really didn’t matter because, if the investors don’t want the founder on the board anymore, they will find a way to eject you.

As I was leaving the board, the stated plan was to try to become cash-flow positive within a year. The new CEO was going to manage to cash and, if we weren’t able to hit our numbers, he would cut costs every quarter. In my opinion, his operational plan to grow the businesses was a rehash of ideas we had already tried, but I gave him and his new leadership team the benefit of the doubt. Perhaps with better leadership, they could achieve the meteoric growth that I was unable to execute.

The new CEO had quarterly meetings with me so I could keep abreast of what was happening with the company. It seemed like we were going to win an extremely large government contract by the end of 2021, which did occur, even though it was a few months late. Sometime in the Summer of 2021, it was decided that the company would sell. The bankers began pitching the street with a ridiculously high valuation of $800m.

What was I up to during this period? In another aborted sabbatical, I took a month or two off to ski, but as Covid started descending on the world, I decided to move to Omaha to lead GrainBridge, a joint-venture between ADM and Cargill. The vision was to use data from the two companies to help farmers become more profitable, an agricultural data refinery. Over the next 18 months, I’m really proud of what the team built. We immediately cut burn, released a product, built a product development machine, and decided that the best way to pursue our mission was to sell the company to Bushel.

While I was celebrating in Omaha, Descartes Labs was in a downward spiral.

An Unnecessary Implosion: Burn, baby, burn

I only got a trickle of information from October 2021 until April 1, 2022, when I received a phone call from an angry board member. The company was about to run out of cash, hadn’t billed a single penny against a massive government contract we had won, had missed numbers every quarter, and was burning an astonishing $2.3m each month. Even worse, I was informed that the acquisition process had been botched and it seemed like the company didn’t have any options.

I was flabbergasted. How could a company that ended 2021 last year with over $17m in revenue be close to folding? I started calling around: investors, current & former employees, and customers that I knew. I went from being flabbergasted to horrified. Everyone–including board members–agreed that the company and the sales process had been grossly mismanaged. Though the details I uncovered are shocking, they aren’t important to this narrative.

The number one job of a CEO is not to let a company run out of money. Burn had been steadily increasing quarter-over-quarter. Cloud costs had ballooned to ~$800k/month. Hiring had continued, despite an inability to deliver revenue. Why did they not cut costs? The reasoning from the CEO to the board was that a RIF during an acquisition process would be bad optics. That calculation was irresponsible and incorrect: acquirers had to consider not only the acquisition price, but the cost of running the company over the coming years. If an acquirer thinks that the cost of servicing the revenue is greater than the actual revenue, that’s a turnoff.

I quickly wrote a plan for the board to reduce costs and reboot the acquisition process. I didn’t relish this idea, since it would be a lot of work, and I was otherwise occupied. However, it was my duty to the company, our investors, and to our common shareholders, that is, all of the Cartesians who were either working at the company or had worked for the company in the past.

The board rejected the plan because they felt like they had bet on the current management team. Even though I believe I could have run a sales process ending in a price that reflected the true value of Descartes Labs, I do understand their logic. In reality, the board should have seen in Q4 of last year that we were going to run off a cliff in 2022 and, if the CEO refused to cut costs, they should have fired him then. Letting the company run towards a cliff of insolvency was a massive breach of the board’s duty.

I continued to watch from the sidelines. With Descartes Labs running out of money and few sales options, Descartes Labs, despite all of its value, was sold to Antarctica Capital for basically nothing. Existing shareholders got only a small percentage of the new company.

Postscript

In the end, I’m still proud of Descartes Labs, a New Mexico-based science spinout with a broad and important mission to better understand our planet through data. We attracted some of the best scientists in the world. We performed remarkable technical feats, many of which translated into enormous value for our customers. Very, very few startups get as far as Descartes Labs did.

That makes the past 2.5 years so bitter for me and all Cartesians. So many people put their trust in Descartes Labs: investing when we were just an idea, leaving comfortable jobs at universities and labs, moving themselves and their families to Santa Fe, committing their best years to the company. Many employees bought their options as they left the company; tens of millions of dollars of their equity went up in flames because Descartes Labs almost ran out of money.

How sad.

To me, the most painful part of the Descartes Labs flameout was that it could have been avoided. The board acted aggressively exactly when they should have had a sober and reasonable plan. When the house was clearly on fire, they seemed to run around screaming about the house being on fire, but worried they’d get burned if they threw any water on it. One can blame the leadership of the company, but ultimately it’s the board that hires and fires the CEO. The unconscionable burn forced a firesale, leaving all shareholders in a worse position.

I’m sure there are many angles to this story and I welcome alternative perspectives. However, even though I’m sure that I got some things wrong, there’s one thing I’m absolutely sure of: this didn’t have to happen and the blame lies primarily on the investors.

I’ll still be rooting for Descartes Labs, even as a minor shareholder. More importantly, I want to fight for the rights of founders and startup employees. If the story of Descartes Labs resonates with you as a founder: if the board intervened in your company and destroyed value, I’d love to talk to you.

I recently reread GEB and got fired up by how brilliantly Hofstadter fuses computation, epistemology, and consciousness. After failed attempts at explaining the book to three of my smartest friends, I decided to write something up.

The problem is that a simple reduction like “GEB is about how complex systems arise from simpler systems” is akin to describing Ulysses as “a day in the life of Leopold Bloom.” More detailed descriptions run the risk of diving into the depth that’s only understandable after having read the book.

This post is a more modest attempt to explain to myself why GEB is important, and focuses on three mental models that have profoundly affected my life: epistemic limits, self-reference, and isomorphism.

If it causes you to read or reread it, then all the better.

Here we go!

The main character of the book is Kurt Gödel, the most important person in the 20th century you’ve never heard of. Gödel is the kind of guy that shows up to his buddy’s 70th birthday with an exact solution to the Einstein field equations as a present. Despite being the greatest mathematician of his generation, he wasn’t stuffy in the least: his favorite movie was Snow White and the Seven Dwarves.

Gödel is most famous for his Incompleteness Theorems, which established limits on mathematics. For the first chunk of the 20th century, mathematicians were obsessed with formalizing mathematics and then proving meta-theorems about those formal systems. In particular, there was a strongly-held belief that for any well-formed formula (a “grammatically correct” statement in math, e.g., A=B is well-formed whereas AA==+B is not), you could use mathematics to decide whether it was true or false.

If you think about it for a second, this makes perfect sense: it seems like you should be able to determine whether any statement is true or false.

Nope! Gödel proved in 1931 that mathematics is not decidable, an earth-shattering result. He proved that there are statements in mathematics, which are true but not provable within the system. Worse yet, it turns out that you can’t build a more powerful mathematical system. Once a system becomes sufficiently complex, there will always be statements which are undecidable. You’re left with a choice: either have weak system of mathematics or accept that there will always be theorems out of reach. A rough analogy to incompleteness Heisenberg’s Uncertainty Principle, which shows that physics makes it impossible to determine both the position and velocity of a particle with exact precision.

Wouldn’t it be nice if every question had an answer? That’s a lovely fantasy, but Gödel shows that there are fundamental epistemic limits to the universe, things that no genius will help us to know, no alien race could teach us, no machine could be built to solve, and no new kinds of mathematics will uncover. How frustrating.

A key feature of powerful mathematical systems (or perhaps, any system that generates complexity…) is that they involve self-reference, that is, they contain ways of talking about themselves. “This sentence is true” is an example. Because self-referential systems can manipulate and talk about themselves, they systems are very powerful and immediately run into fun paradoxes. Is the statement “This sentence is false.” true or false? Either way, it doesn’t end well.

A third major theme of the book is isomorphism, which is unique to Hofstadter’s vernacular. In formal mathematics, “isomorphism” takes on a version of “equivalence.” For example, it turns out that many different formalizations of mathematics are provably isomorphic, like Turing Machines, arithmetic, set theory, and formal logic. Hofstadter deliberately uses the term more loosely to describe two systems that are structurally similar. I find this quite useful because it forces one to define the structures of the system, why they are similar, and why other parts of the system are less important. We might describe the way that planets fly around stars as isomorphic to the way that electrons fly around nuclei.

The two minor characters, M.C.Escher and Johann Sebastian Bach, are reflections of Gödel in art and both liberally use self reference. Escher draws pictures of hands drawing hands (!) and water “falling” in an infinite loop. His images don’t just play tricks on the eye, they force paradoxical conclusions, regardless of your angle of interpretation. On the musical side, Papa Bach was most famous for his complex fugues, which are basically the same melody played on top of each other. Common versions of this you might have sung as a child are “Row, row, row your boat” and “Frère Jacques.” Both Escher and Bach are woven into the story (like a fugue?), providing tangible examples to the more abstruse mathematical concepts.

Perhaps the most astonishing part of the book is the quality of the writing itself. Each chapter begins with a clever dialog between Achilles and the Tortoise (inspired by Lewis Carroll), and a few of their anthropomorphic friends. They deal with a whole range of bizarre situations, like record player so powerful that it can play any record (including a record that can destroy the record player) and asking a Djinn for a meta-wish (“I wish for 5 more wishes”). Hofstadter’s greatest achievement is his palindromic Crab Canon in Chapter VII, which is a dialog that can be read backwards and forwards. Of course, these aren’t just cute dialogs: each is isomorphic to the themes in the following chapter. Oftentimes a dialog is a more understandable exposition of the chapter’s theme than the chapter itself.

And, naturally in a book about self-reference, GEB itself is highly self-referential. Themes are often resolved hundreds of pages later and require going back to appreciate fully the depth of Hofstadter’s argument. Mercifully, he’s a gifted and lucid writer so, even though there are chapters that are dense, it’s always tractable to read.

After 742 pages and even after having written the paragraphs above, I still struggle with a simple answer to the question: “What is this book about?” The best I can come up with is that GEB equips you with mental models to contemplate philosophy.

So to end, a few personal examples about how GEB has influenced my own thinking.

I recently joined Stand Together, who shares my strong belief in bottom up solutions. Perhaps the idea that bottom up solutions are better isn’t just an empirical statement of sociology, but fundamental to the nature of complex systems. Indeed, Hofstadter goes through many examples of how complexity emerges from simpler systems, often which look nothing like the higher-level systems. Consciousness itself doesn’t exist in neurons, and yet neurons as a system create consciousness in humans (this is critical to Hoftstadter’s argument that machines can think). There’s also a fantastical example in a dialog with the Anteater, who has conversations with Aunt Hilary, an ant colony. She is perfectly capable of having a robust conversation with the Anteater, powered by the ants in the colony. Of course, the ants themselves are individuals with their own cares and concerns and has no knowledge of the emergent intelligence, much like Aunt Hilary has no knowledge of her inner workings.

How DNA expresses as proteins, how the brain functions at multiple levels, how we understand and use words, how programs don’t have access to the underlying transistors, how Aunt Hilary doesn’t know what the ants are doing… all of these are a set of isomorphisms that suggest that bottom up is better than top down. An additional tenet of Stand Together is “believe in people,” which means that the smallest units act intelligently. Like ants or neurons, we make local decisions every day that bubble up into the structure of society, without anyone telling us what to do.

The idea that epistemic limits exist in something as universal as mathematics has humbled me about the limits of knowledge for complex human systems. Utopian thought experiments often generate useful frames of exploration, but ought not be confused with reality. Utopians often try to pull out the “bugs” from human systems, often which are endemic to the system, or “features,” as we might say in the trade. Bugs might not be desirable, but sometimes, the bugs can’t just be ripped out of the system without destroying the system itself. Our time would be better spent figuring out—within the system—optimize for minimizing the downsides of the “bugs” while maximizing the value of the features. Think about this with respect to capitalism, socialism, and communism…

A final area where GEB has influenced me is in designing software products. Hugh Dubberly has been my collaborator for years, starting off with our deep dive into cybernetics, the study of feedback loops. We believe that iteration is key to quality; perfection is impossible out of the gate. Further, the system used to generate quality software is a series of feedback loops between customers and the company, product and engineering, and so on. Though specific product frameworks have changed over the years, that obsession with iteration and feedback permeates everything I’ve implemented.

My modest goal in writing this post was to have something I could send to a friend, rather than to spend an hour fumbling a feeble explanation of Gödel, Escher, Bach. I had a secondary goal in the back of my head… if you have a copy of GEB on your shelf collecting dust and you’ve never read more than a chapter or two, dust it off and see how it goes this time.

Today is my first week as the inaugural CTO of Stand Together, where I’ll be on a mission to operationalize the idea that “our principles are our product.”

Stand Together is a philanthropic community tackling the root causes of the biggest problems facing humanity. How do we do this? Our over 700 business leaders and philanthropists support over 200 community-based organizations like The Phoenix, which is addressing addiction through a fitness community. We take a systems approach to solving issues, bringing together policy, communities, and scholars. For example, with criminal justice reform, we’re proud of the First Step Act, legislation that was passed in 2018 by an overwhelming majority during an extremely divisive time in our country. While exciting, the Act was a first step and we’re not even close to being done.

Our approach is based on first principles: we believe in people, we recognize bottom up solutions as the best way to tackle problems, and we will unite with anyone to do right. As our founder, Charles Koch, reminded us recently on a podcast, the core principles of Stand Together are the principles of human progress, enshrined in the Declaration of Independence. When we adhere to those principles, society flourishes; when we don’t, society suffers.

Being the first CTO, I’ll spend my first few months unpacking the role, but I do have a few observations to make.

As Marc Andreessen famously stated: software is eating the world. Similarly, data and models will run the world. Every organization, even a philanthropy like Stand Together, will succeed only when it elevates technology as a core part of its strategy and culture. Technology is an agent of change.

Zooming out, we have an opportunity to consider how technology will affect the key institutions of society — government, education, community, and business — and to dream about an infrastructure that will scaffold and help scale the work of our partners.

On a personal note, I’m excited to dive into these monumental questions because they align with my own philosophical interests. My plan after exiting my last company was to take a sabbatical and go back to the basics in economics, philosophy, and technology. What can economic thought teach us about value, exchange, and incentive systems? Why are the esoteric philosophies of the post-Marxists so popular and showing up in major papers? And what can we learn about social media from previous major media transitions like the printing press or the telegraph? Those three threads are linked by technology (an exercise left to the reader) and I can’t wait to dive deeper.

I’m not entirely new to Stand Together; I’ve been an entrepreneur partner and donor for the past few years. In Stand Together, I’ve found an organization that reflects my values and is willing to ask pressing philosophical questions and embrace the idea that no matter how well we may be doing, we can always do better.

I share a belief with Stand Together that we learn by doing, so very early on I’ll be looking for a product to build. For example, a key component of our support for social entrepreneurs is to help them apply Market Based Management to improve their organizations and overall performance. What if we could turn that management philosophy into software that empowers better operations? Further, data has the potential to create positive feedback loops that make organizations more responsive to the needs of its customers. What if it could help philanthropies to better measure their impact using data and models? I expect that we’ll uncover a number of projects where software, data, and models will help us accelerate change in the world.

I started on Monday, so I’m already a few days in. So far, I’ve been blown away by my colleagues. They are incredibly kind, intelligent, and thoughtful about creating change in the world. I’m honored and humbled to be part of this incredible organization.

I’ll be moving from Omaha to DC over the next month, so do reach out if you’re in the area.

While I was in Los Angeles, soaking up some sun and visiting with space companies (plus clocking my millionth mile on United), AgTech was having a big week.

Bayer+Microsoft announced that they would partner up to build “digital solutions” for agriculture.

My instinct, like Louisa Burwood Taylor at AgFunderNews, is to ignore press releases about “partnering” with a cloud provider. It’s akin to a company announcing that they’re using Slack to improve internal communications. Nope: you’re actually a customer, just like everyone else.

Let’s assume though, that this is something very real and and that Bayer is building out a digital platform for building better software in agriculture. Because that would be a very interesting announcement, indeed.

What is a digital platform? It’s a set of software, services, and data that make it easier to build software. iOS is a platform because it contains a development environment (where you write the code), lots of UX widgets (like buttons), services (like Apple Pay), and many other out-of-the-box features. If you had to build all that from scratch, it would take a lot longer to build an application. Key to a platform is that you’re only a platform if others build on top of you.

Should there be a digital platform in AgTech? Whatever would it do?

I certainly can imagine some shared data services. FieldView already has the ability to integrate with other pieces of software to share things like field boundaries, perhaps that could be part of the platform. External data, such as USDA data or NOAA weather data could be very valuable.

I can also imagine taking the data refinery approach, and combining sets of data to create standard services. Satellite data could automatically identify fields or a crop production forecast could be built from weather data. That’s a tough, crowded business; I should know, as a cofounder of Descartes Labs.

Perhaps there’s a much, much bigger play here.

Seana Day on Twitter: "I kinda wish someone would do a tech stack visualization so we could really unpack this. Kidding, not kidding. But what makes me super curious is how MSFTs more recent moves in the #agtech space start to converge with their huge GP / Dynamics footprint in Ag. pic.twitter.com/UVQ8ktmN4l / Twitter"

I kinda wish someone would do a tech stack visualization so we could really unpack this. Kidding, not kidding. But what makes me super curious is how MSFTs more recent moves in the #agtech space start to converge with their huge GP / Dynamics footprint in Ag. pic.twitter.com/UVQ8ktmN4l

Great Plains and Microsoft Dynamics (i.e, enterprise software) have a big footprint in agriculture. So does Microsoft itself as an OS and IT platform, so much so that I still can’t do a Google Meet with most ag corporations: it’s all Teams. Perhaps this is a venture into moving its cloud business into agriculture, to make sure GCP can’t get a hold on these companies and lose valuable Office contracts to Google Docs.

Building a digital platform targeted at big agricultural companies could be a perfect solution, as they’re all struggling to become more innovative and digitally nimble.

Imagine a development environment where all the data is easily available(satellite, weather, price, etc.), there are widgets for all kinds of agronomic analysis, functions like payments are standardized, and there are integrated tools for communication (e.g., text messages). Then, instead of a massive investment in expensive, scarce engineers from MIT, ag companies could focus their development resources on the product itself.

I have trouble believing that this platform company could exist within a drug and chemical conglomerate; but if they truly believe that digital is the future of ag, why not try to create the platform that runs the whole thing?

Let’s check up on it in 6 months.

Second, FBN announced another huge round of funding. Last time they raised money, I suggested that they buy up AgTech startups. Instead, they followed a wiser path, focusing on their customers and their core business. They continue to hint at an IPO, though these days it seems that anyone can IPO (see Farmers Edge’s laughable Q3 results; and resulting stock crash).

However, I can’t find anyone though who really understands FBN’s business. Any good perspectives? Thoughts?

Finally, GROWMARK+CHS announced a venture fund for breakthrough AgTech. What’s most interesting here to me is the following line:

“The combined markets of the two companies cover millions of acres and thousands of farmers that will create an industry-leading test field for products and services within North America.”

Capital can come from anywhere; access to potential customers is absolute gold. I’d love to see the two cooperatives create a beta technology platform to distribute to farmers at scale. One of the biggest problems in getting an AgTech company off the ground is farmer adoption. With a network of farmers primed to try out new technology, that could be a real accelerant for portfolio companies.

Agriculture is the biggest industry that Silicon Valley has failed to disrupt. Despite billions of dollars going into high-flying AgTech software startups like Farmer’s Business Network (FBN) and Indigo Ag, there hasn’t been a single industry-changing success since Climate Corp a decade ago.

A favorite model of the biggest AgTech companies is the hallowed marketplace, the great disruptive tool of the digital era. Think Amazon, eBay, Kayak, AirBnB, and Uber. In agriculture, all such attempts at building a marketplace have failed miserably.

I declare October 12, 2021 as the day that the grain marketplace died.

Why? Because today, we announced that GrainBridge has been acquired by Bushel. Not only will Bushel inherit GrainBridge’s team and data refinery, it also secures two customers, ADM and Cargill, the two largest grain buyers in the US.

Bushel’s business model isn’t a marketplace. They are creating the digital infrastructure for the agriculture industry. Instead of trying to disrupt, they’re bringing agriculture into the digital era.

I don’t come from an agriculture background. It’s taken me over a year at the helm of GrainBridge to figure out why agriculture is so resilient to disruption and why marketplaces aren’t tenable. I’ll try to unpack the bombastic statement above by first teaching you some key points about grain markets. Then, we’ll use that knowledge to show why grain marketplaces don’t solve problems and finally, how Bushel represents an alternative vision.

A primer on grain markets

The term “agriculture” is unfortunately broad. For the purposes of this article, I’m talking about grains in the US: corn, soy, and wheat being the three biggest crops. The bulk of these row crops are in the midwest. For example, Illinois and Iowa produce about 26% of soybeans and the top four corn states produce 56% of total production. Corn is grown in about equal parts to produce sugar, ethanol, and feed; soy is grown mostly for feed and the oil it produces; wheat is grown mostly for food. Each of these crops is a commodity. That is, corn produced in a field in Iowa should be the same as corn produced in a field in Nebraska. Importantly, grain is exported. The US exports about 50% of its soy crop. The US agricultural system literally feeds the world.

With tens of billions of dollars of crops grown, you can imagine VCs and entrepreneurs drooling a massive Total Addressable Market (TAM). Marketplaces have disrupted other traditional industries, why not build one for agriculture?

To understand why that won’t work, we need to understand the basics of farming.

Hundreds of thousands of individual farmers across America own their own business. They lease or own their land, manage their equipment, buy inputs (seed, fertilizer, etc.), and sell their grain. Unlike speciality crops like strawberries, where buyers will often contract with growers, grain farmers choose what crop to plant independent of there being a buyer and will sell it at the time of their own choosing.

Since grains are a commodity, the price fluctuates on national commodities marketplaces, like the Chicago Board of Trade. These futures contracts are a promise to deliver grain at some point in the future. Farmers watch the prices and try to sell when the price is high. This is called grain marketing, a term of art for a farmer selling his grain or… “bringing it to market.”

The farmer sells to a local storage and distribution facility called a grain elevator (though sometimes it’s a processing facility, like an ethanol plant). Transporting grain to a facility has a transportation cost, so there are only so many buyers within a reasonable range, limiting potential buyers to any given farm to a ~35 mile radius.

The buyers at the grain elevator need to make money, so they set a price for storing and handling the grain, called basis. Typically negative, the basis is added to or subtracted from the futures price to arrive at the price the farmer gets paid, the cash price.

A farmer typically has grain storage, called bins, on his farm. That means that he doesn’t need to harvest his crop and take it directly to the grain buyer. Harvest is often the worst time to sell grain, because with supply very high, prices will typically dip. So, the farmer can enter into a contract to deliver later on, say January 2022, when the price might be higher.

And, it turns out that each of those delivery months has a different price that fluctuates independently. Now things are really getting complicated. How these prices fluctuate doesn’t really matter. The point is that a farmer — while he’s trying to tend to the other parts of his business, like growing his crop — also has to worry about corn prices, which move due to everything from monetary policy to trade wars.

I only scratched the surface of grain marketing and simplified some of the concepts. Don’t worry if you didn’t make a lot of sense to you, even farmers who have been selling grain all their life can find grain marketing challenging.

For the rest of the discussion, there’s really only two things you need to know:

1. There are only so many possible buyers in a region, due to transportation costs.
2. There are only so many farmers in a region and a grain buyer likely knows all of them.

Grain marketplaces: solving the wrong problem

The first grain marketplace, Rooster.com, was launched with fanfare back in 2000 during the original internet craze. In reading this article from the WSJ upon launch, a lot of the theory made sense. After all, farmers are rural and it would be hugely advantageous if they could conduct business online. Change a few details in the article and you could probably raise a seed round with that business plan today. Rooster.com sadly succumbed to the rash of internet startups that died after September 11, so we never know if it would have evolved into something else.

Fast forward to today and there are a whole number of marketplace style startups. Some, like FBN, have expanded from their original marketplace vision and are trying to create a comprehensive experience for the farmer, from crop inputs to farm management. Others, like Indigo Ag, built a marketplace, made some missteps, and is doubling-down on the marketplace as per this recent article by Jacob Bunge at the WSJ.

I wish all of these companies the best of luck — lord knows that I’m not always right in my prognostications, but I would like to call out a huge cautionary note: marketplaces don’t solve problems for the farmer or the grain buyer.

On the farmer side, you may already anticipate where I’m going from the explanation of grain marketing above. The farmer only has so many local buyers and the price is mostly determined nationally. The farmer’s biggest problem is making a decision: when’s the right time to sell grain? Picking the right time could be the difference between squeaking by and having a great year.

In fact, that’s exactly why ADM and Cargill decided to form GrainBridge back in 2018. They wisely saw too much venture capital going into marketplaces and wanted to help farmers become more profitable. The idea was to pool data about farmer selling behavior into GrainBridge and build a decision support application that would give farmers unbiased advice on when to sell grain. The platform would then attract other grain buyers, creating data gravity, and ultimately being a necessary tool for the farmer to make smart grain marketing decisions.

On the buyer side, marketplaces theoretically give you access to new customers. That’s the power of something like Amazon for a small vendor. But, a grain elevator typically knows all of the local farmers. They’re not going to magically discover a massive farm that they didn’t know about. Their challenge is how to stay in touch with farmers and to gain more visibility into when grain is going to arrive at their facility.

So, neither buyer nor sellers want a marketplace. What do they want?

That’s where Bushel comes in. Digitization has always been the challenge of AgTech. Most grain buyers are running on ancient grain accounting systems that weren’t meant to be on the internet, some of them running on hardware that would have been considered antiquated in 2000, let alone in 2021. That meant that many grain buyers had trouble digitizing the basic transactional documents between them and their farmers. And that’s where Bushel came in.

Digitizing infrastructure isn’t sexy, it’s really hard work. There’s a huge payoff though if you can create the digital platform for innovation in AgTech. Leaf’s G. Bailey Stockdale last week wrote a great post about the interplay between apps and infrastructure. My response on Twitter was that I don’t see enough companies focused on platforms in AgTech.

Mark Johnson on Twitter: "#AgTech is the worst of both worlds: products that don't solve a clear & painful problem for farmers/the industry and very few companies working on building a platform for further innovation.Digital infrastructure is necessary to build anything interesting. / Twitter"

AgTech is the worst of both worlds: products that don't solve a clear & painful problem for farmers/the industry and very few companies working on building a platform for further innovation.Digital infrastructure is necessary to build anything interesting.

I believe Bushel’s hard work and sweat put into digitizing this formerly paper processes will be the foundation that launches them to become the most innovative company in AgTech. (Oh, and ICYMI, Bushel raised $50m earlier this year and added a few major customers.)

A parallel is Amazon versus Shopify. Grain marketplaces have been trying to build the Amazon of grain for a long time, while Bushel quietly built the Shopify of grain. Every grain buyer wants their own storefront, their own customer relationship management, and their own analytics. That kind of customization is where Bushel shines. Each grain buyer can create a customized experience for their farmers; and farmers now have a standardized digital interface for what used to be a paper process. Everyone wins.

Walt Duflock once asked why there isn’t an eBay in agriculture? My answer is: we don’t need one.

Farmers have a tough job. Not only do they spend time in their fields maximizing the output of their crops, but they also need to pay attention to commodities markets, figuring out when to sell their crop.

At GrainBridge, our job is to help farmers make better decisions about when to sell (or “market”) their grain. We believe that data is the answer, which is why we undertook the task of building a data refinery.

The idea behind a data refinery is to ingest multiple datasets, clean them up, and put them in an environment where the data is easy to access by everyone in the company, whether it’s a PM who needs user data or a data scientist building a predictive model. We also needed to build the tools to serve that refined data back to our farmers.

Our legacy with data goes back to our founding. As a joint-venture between ADM and Cargill, the two largest grain buyers in America, GrainBridge has a huge dataset of grain transactions in the US, going back several years. By combining those data with external datasets like historical futures prices and USDA market forecasts, plus adding in our own user data, we can then build decision support to help the farmer make smart, profitable decisions.

Our first application, launched in September of last year, gave farmers digital access to their data. The architecture of our app is largely transactional, focused on scaling up to millions of farmers, storing hundreds of millions of records, and ingesting data very quickly.

For the data refinery, we needed to build a new architecture, more tuned towards analysis, where we can scan through millions of records, rather than just return a set of records associated with a single farmer. We also needed to be prepared for any kind of large dataset, structure or unstructured, from time series, to weather, to satellite data. We don’t yet know all the data we need, so flexibility was important to us.

Ultimately our goal is to use data to become a personalized AI consultant, giving the farmer suggestions to help maximize profit and reduce risk. However, we started off with a straightforward MVP.

GrainBridge released our first data-focused feature in Q2. Instead of rushing out with minimal infrastructure built on epoxy and duct tape, we decided to spend time building out the basic pieces of the underlying data refinery, so we could accelerate our experimentation for the rest of 2021 and beyond. That way, we could accelerate data experiments for the rest of the year and transform GrainBridge into a data-animated organization.

I recently wrote about how GrainBridge became a product development machine. Now, I’d like to share some lessons about how we put that machine to work and built an agricultural data refinery.

Hire data expertise…

One of the first hires after I arrived at GrainBridge was a top data scientist with experience building models on top of large consumer behavior datasets, plus several years at a big data platform. Brian has been the data inspiration for the team, sharing his wisdom based on a career of successes and mistakes. He regularly scheduled presentations to the team, outlining his vision for building a data refinery and what we could do with it.

In addition to Brian, we added another data scientist, a data engineer, and have an open position for a data BA (product support for data).

We also leveraged our relationship with AWS. We leaned on them for architectural support, to make sure that we were making sound, long-term decisions.

…but, everyone on the team should understand data

If your company’s product is based on a data refinery, then everyone in the company should understand data.

Spearheading our internal effort was GrainBridge’s long-time architect, Fayez Barbari (who sadly has left the company). He had spent months ahead of the project educating himself about big data technologies, building prototypes, and creating a prototype architecture. Together with AWS, he ran a data workshop to train the entire team on the AWS architecture we’d be using and the related programming paradigms. Even though everyone wasn’t working on the data refinery, we felt it important to have everyone understand it.

That’s helped tremendously both in building the data refinery and having product conversations about what’s possible with our new architecture.

Spend time understanding the context of the data

The data we received from ADM and Cargill included contracts (what kind of grain to be sold, where, at what price, and on what delivery date), delivery tickets (when grain was delivered and what quality), and settlements (payment after the fulfillment of a contract). However, each of these pieces of data has dozens of fields. Some of those fields, like the type of contract, have dozens of different possibilities.

Though it was time consuming for our owners to explain the business and operation context of the data to us, this was a critical step. It isn’t enough just to drop data at the feet of a skilled data scientist and expect them to stumble on gold. ADM and Cargill provided insights and contextual information that is simply impossible to extrapolate from the data alone.

Create a model of your solution space

The worst direction you can give a data scientist is “see what you can do with this dataset!” A much better, user-focused question is: what data features would be valuable to an end-user and are those features possible with this dataset?

To focus our thinking, our data scientist came up with a model very early on:

• Score — giving every farmer a score related to some aspect of their grain marketing
• Benchmark — for any given score, how well does the farmer do with respect to others
• Insights — how do farmers who consistently score high relative to various benchmarks market their grain compared to those who do not, and can we communicate those insights to others

That framework gave us a “solution space,” a framework to discuss the types of features that we could engineer on top of our datasets, which ultimately led to…

Have a clear MVP goal

If I had to choose one key lesson, this was it.

For our MVP, we were not optimizing for user value, rather, we were trying to optimize for something that was the simplest possible feature that allowed a roundtrip of data through the data refinery. Of course, we also wanted something that was useful to our farmer customers, but it didn’t necessarily have to be the most valuable feature for them. It’s more important to get something out to test the infrastructure.

We decided to choose the average sales price of the contract. It was something that was easily calculable given that we had contracts for each of our farmers. And it was a useful number that they probably wouldn’t have ever calculated for themselves. Plus, given the data feature model above, you need a score if you’re going to benchmark farmers against their peers or breakeven price.

Build out infrastructure for the future

A key principle was that we didn’t just want to build our first data feature, we wanted it to be built on a working data refinery. This is a tough balance to maintain: on one hand, we didn’t want to write custom infrastructure for a single feature, on the other hand, we didn’t want to build out the complete data refinery so that any feature could be built. Plus, security and data privacy are critical to us to maintain the trust of our farmers, so we couldn’t just slap something together.

Our head of engineering went to the white board and sketched out the complete data refinery. We used that model as our basis for building a “round trip” with our MVP feature: could data be ingested into the data refinery, analyzed, and then appear in the final product. In this way, we were still creating pieces of each of the pieces of infrastructure, while still focused on delivering customer value.

What’s next?

Our first order of business is to finish all of the components of the data refinery that need refining. There are more datasets to ingest and more tools to build. We’re busily working on that this quarter, with the hopes of releasing at least one more public data-focused feature.

Even more exciting is that we can start asking questions from our data: what makes a farmer really good at selling his grain? What are the habits of farmers who aren’t as profitable? Based on your past habits and the market, do we recommend that you sell today or hold off until the future? Are local prices historically really good or really bad? How does your average price compare to other farmers like you?

These kinds of questions will allow us to get closer to our goal of helping farmers to become more profitable.

What about your company? Have you built a data refinery? What have you learned?

Of course, visions need to be translated into reality. My charge was to transform GrainBridge from an agricultural company (that built software) into a software company (in agriculture).

During my first week, the team told me that we wouldn’t be able to release a product until 2022. With a broad, complex product roadmap and no data yet from our owners, the road did seem long and daunting.

I challenged the team to release the simplest product as quickly as possible because I have a deeply-held belief that great products are built only through iteration. Finding product market fit with the waterfall method is just too risky. No matter how much user research you do, nothing beats having real users give you feedback on real software.

Fast forward to a year later and I couldn’t be prouder of the team here at GrainBridge. We released our first version of the new GrainBridge in September of last year and have been consistently releasing every 2 weeks since then.

How did we do it? We had a relentless focus on process.

Luckily, from a software perspective, we weren’t starting from zero. Before I arrived, the team had built a robust, modern infrastructure on AWS, capable of ingesting millions of records of data and able to scale to millions of users on the front-end.

With that foundation, the first order of business was to distill our product plans into a reasonable PRD. We asked ourselves what it would take to release by the end of summer, forcing us to evaluate whether every feature was necessary for a minimum viable product. Even though that meant there was no fancy data science in v1 of GrainBridge, it helped us to scope the product down to under 6 months of development time.

GrainBridge launched our initial product in September. A farmer could connect to her accounts on ADM and Cargill and see all of her contracts, scale tickets (deliveries), and payments. As far as we know, this was the first time transactional documents were aggregated automatically from two grain buyers on the same platform. GrainBridge was far from perfect, but that wasn’t the point of the release. The important thing was to release, gather feedback from our customers, and iterate.

The next step after the initial release was to move to regular release cycles. After a lot of internal debate, we settled on an aggressive two-week sprint cycle. Unsurprisingly, our first few releases were horrendously painful.

Product was behind on stories, engineers were spinning wheels, our metrics didn’t give us any information, and everyone was extremely frustrated… and yet, we persevered. By the time we hit the beginning of 2021, we had our first solid release and every sprint in Q1 was released on time, many of them at a faster velocity than we had predicted.

This process wasn’t easy and we certainly didn’t execute perfectly. But again, that’s OK. Product development process is just like the product: it only gets better through iteration.

Our best tool was the weekly retro. At first, retros were quiet. The real conversations would happen afterwards. Eventually, people realized that speaking up in the retro — voicing the problems that were slowing us down — was the only way to bring them to light and solve them.

Becoming a product development machine in under a year is a feat that few companies can match. With all the work we did in Q1, I can confidently say that GrainBridge is a best-in-class product for farmers who want to view their contracts, scale tickets, and settlements and keep up on local cash prices from ADM and Cargill.

GrainBridge has a much bigger vision of helping farmers to know when to “market” (that’s agriculture-speak for “sell”) their grain to help them to become more profitable. Now it’s time to point our product development team at that problem.

Our main advantage is that we have years of data from the two largest grain buyers in the US. Those data can help us give insight to the farmer — all done safely and securely, with respect for privacy. We have a notion that we’ll be able to analyze our data to figure out the best marketing strategies to maximize profit. Internally, we joke that we’ll be able to boil it down to the “Seven Habits of Highly Effective Grain Marketers.” Once we know those attributes, we can give each farmer a grain marketing score and how they might consider improving.

We’re building out the foundations of a data refinery right now to build and test out user features. Our data science team now has two PhD data scientists and a data engineer. If you’re interested in joining, we have an open business analyst (BA) position focused on data.

Finally, the main point of building a product development machine is to solve problems for our farmer customers. In Q1, we conducted ethnographic research (i.e., observing the farmers in their environment) on a group of farmers. Grain marketing software has never been widely adopted in the industry. Our biggest insight from the research is that adoption is driven by trust; and it’s much harder to trust a piece of software than a human being. We’ve got some ideas about how to build up that trust with the farmer and lead him to higher profits.

GrainBridge has achieved so much in this past year thanks to every member of our small strong team, the incredible support we’ve gotten from our board and owners, and the feedback we’ve gotten from our customers. I can’t wait to report out on some of the releases we have coming up.

Agriculture is responsible for about a quarter of global greenhouse gas emissions.

Given the scope of the general problem of sustainability in agriculture, I’ve been researching how to address carbon emissions at the farm level. Since the market is focused on carbon sequestration, that’s where I’ve focused my attention.

Why is carbon sequestration such a hot topic? My working thesis is as follows:

• There’s no silver bullet for sustainability in agriculture.
• We must focus on solutions that are effective and easy-to-adopt within the current system.
• Farmers are driven by ROI.
• Land-use change is a major contributor to agricultural emissions.
• Plants remove carbon from the atmosphere.
• More science needs to be done to understand carbon capture.
• We need a ledger to track emissions and a marketplace to buy and sell credits.
• There is a lot of innovation in this space.

This post is a combination of my own research and a reflection of a the conversation we had last week in our Entrepreneurs in #AgTech Clubhouse, which will be continued this Friday. I’d welcome comments from industry experts, companies working to address sustainability, and, perhaps most importantly, the perspective of farmers themselves.

There’s no silver bullet for sustainability in agriculture

In the world of energy, it’s generally agreed that ultimately, we need to stop pumping hydrocarbons out of the ground. Folks will disagree on the timeline — whether it’s going to be decades or tomorrow — but the idea of creating cheap (essentially free), portable, zero-emissions energy, is a clear end goal. Technologies like nuclear, solar, and wind, combined with advanced batteries, are a framework for achieving zero-carbon energy.

Unlike energy, the sustainability response from agriculture is murkier.

Plants are wondrous little machines, utilizing the natural resources of the planet — sunlight, water, nutrients in the soil — to create the sugars and proteins we humans need as energy.

Plants aren’t going away any time soon, so we need to figure out how to grow enough food for the planet while doing so as sustainably as possible.

We must focus on solutions that are effective and easy-to-adopt within the current system

In the above graph, the x-axis is the ease of adoption and the y-access is the impact on greenhouse gas emissions.

Dreaming about big solutions may be fun, but often impractical. For example, livestock is a large source of emissions, but converting everyone on the planet to vegetarianism is a fantasy. Though the potential of lab-grown meat from companies like Impossible Foods and Beyond Meat is exciting, we need solutions for today, not decades from now.

How do we find solutions in the sweet spot: relatively easy-to-adopt and high impact on sustainability?

Farmers are driven by ROI

Farmers are the ultimate entrepreneurs, betting every growing season that through a combination of grit and technology, they can bend Mother Nature to their will and grow food to maintain their business.

If we can find solutions that drive positive ROI on the farm and benefit the environment, then we’ll have a far better chance of adoption at scale.

An example of this is how seed technology has improved crop yields in the latter half of the 20th century. Farmers adopt new seed varieties because it increases their yield, improving their revenue; and it means that we need less total farmland to grow more food for the planet.

Land-use change is a quarter of agricultural emissions.

The most striking fact (to me) in my research is that about a quarter of agricultural emissions is from converting natural biomes into farmland. Plants suck carbon out of the atmosphere and store it in themselves (like trees in a forest) and in the ground. Disrupting these processes means less carbon is sequestered naturally.

That means that if we want to have a meaningful dent in carbon emissions, we can’t just focus on the agricultural process, we need to have a long-term path to converting some percentage farmland back to nature, whether that’s forest or tallgrass prairie. That’s a tall order, likely decades in the making.

Luckily, there might be ways to have our cake and eat it too, with a combination of farming practices, including cover crops and soil management.

Plants remove carbon from the atmosphere.

Humans require oxygen to create and return carbon dioxide. Plants are the opposite, and pull carbon dioxide from the atmosphere and return oxygen. We can use this fact to encourage farming practices that maximize the amount of carbon captured in the soil. For example, if we can encourage cover crops and tilling practices that leave soil undisturbed, there’s the potential to store a huge amount of carbon within existing farmland.

More science needs to be done to understand carbon capture.

Like any cutting-edge science we need a lot more data.

How do we determine how much carbon is being captured? Today, the state of the art involves pulling out multiple soil cores from a field and baking them in an oven. That process is not scalable. Plus, it doesn’t account for the variation in soil types (300,000+) across the US. We’ll need a lot more data so we can accurate estimate how much carbon is actually being sequestered by a farmer.

There are even skeptics who suggest that we’re not capturing nearly as much carbon as we think or creating new problems. For example, some research suggests that increased carbon levels breed microbes that belch out nitrous oxide, an extremely pernicious greenhouse gas. Even if the carbon capture thesis is correct, it will take some serious science to allay all of the concerns.

The worst case scenario is if we put a ton of effort behind carbon capture and find out that we didn’t capture as much as we had hoped.

We need a ledger to track emissions and a marketplace to buy and sell credits.

A carbon credit market works by one party selling carbon credits and the buyer using those offsets against their own carbon output. In order to do that well, it’s critical to track how much carbon is sequestered by a farm, which needs to happen over time. That’s exactly why the science becomes so important.

A challenge here is that the bad taste of the carbon market collapse in 2010 is still felt; and farmers have not shown willingness to enter into speculative markets, even if the promise of additional revenue is there. Less than 1% of farmers have entered a carbon contract, according to this survey.

There is a lot of innovation in this space.

Trace Genomics is trying to better understand soil. Indigo Ag is focused on a grain marketplace with their carbon offering being a centerpiece. Nori has a carbon removal marketplace. Truterra is Land O’Lakes’ ambitious carbon project. And many, many more.

On one hand, all this innovation is great. Venture investors and large companies are putting a lot of dollars behind carbon marketplaces. On the other hand, there is still a lot of noise, which makes it difficult for a farmer to decide to participate in something so new and unproven.

Conclusion

My conclusion is that carbon capture is a fascinating area with lots of promise, but more research needs to be done before we all jump on the bandwagon; plus we’ll need agreement from key actors in the supply chain before it will be widely adopted.

What am I missing? What companies are innovating in carbon? Is there something beyond carbon capture I should be looking at? Where do you see sustainability going in ag?

Interested to hear more from industry experts. We’re actually hosting a second Clubhouse this week because last week’s discussion was rich, but inconclusive. Do join if you can.

A reading list

• Why Carbon Capture hasn’t saved us from Climate Change Yet. (538, 2019)
• Farms Can’t Save the Planet (TNR, August 2020)
• Can ‘Carbon Smart’ Farming Play a Key Role in the Climate Fight (Yale, March 2020)
• Could Paying Farmers to Capture Carbon… (Mother Jones May 2020)
• Environmental Impacts of Food (Our World in Data, January 2020)
• Death to the Chicago Climate Exchange (CCX) (Nov 2010)
• 80 Million Reasons not to Pay for Regenerative Farming (Sarah Mock, Feb 2021)