Check out the conversation on Apple, Spotify, and YouTube.
OpenAI’s structural failure (2:57)
Aakash: Eric, I think the structure that is on the top of everybody’s mind is OpenAI. OpenAI started as a nonprofit. It created a capped profit arm. Then they just converted to a for-profit PBC. Based on everything you just showed us, what does that structural trajectory tell you?
Eric: OpenAI is a really funky example. It’s kind of hard to know what lessons to draw because it’s such a singular and strange story. OpenAI has been anything but stable. Two or three crises that have almost ended the company.
OpenAI, for all of the attempts they made to protect the mission, never had this structure. It only ever had one board. And I think that was a big mistake.
The board naturally had the responsibility to hire and fire the CEO. Sam used to brag about that. He would say, the reason you can trust me is I’m accountable to a nonprofit board who can fire me anytime. But when they actually tried to fire him, they did it in the middle of trying to close a tender offer that he had personally negotiated.
All the employees were about to make millions from this tender offer. Microsoft hadn’t been given notice. The other investors were all tied to Sam. So although the board had power on paper, the actual power was held by the employees, the supplier, and the investors.
Governance is a study of power relationships. Every organization reflects somebody’s vision of utopia. When they grow, they turn the outside world into themselves. Amazon was once described as a machine for creating more Amazon. That’s what every company does. If it has integrity, it makes more of the world the way it thinks it should be.
Dario called Eric early (6:27)
Aakash: One of the craziest stories you tell in the book is that Dario Amodei gave you a call in the early years of Anthropic. When you analyze their governance structure, are they implementing things in a better way?
Eric: I think so. Of course I played a role in doing it, so of course I think it’s great. People misunderstand this all the time. I am not taking credit for Anthropic’s success. All credit to Dario and that exceptional team for what they have done. Nor am I saying they are perfect.
I met them right after they had left OpenAI. They were still figuring out the seed round. They were very concerned with their ethos of AI safety. Their actions speak louder than words. They have put their money where their mouth is.
They said, we want to protect this ethos as we scale. They understood this was going to be a multi-trillion dollar company if it worked, which meant the temptation to steal the technology or misuse it would be immense.
To do that required them to defend this idea for two years as they raised their Series A, Series B, and I think it was in the Series C they finally enacted the Long-Term Benefit Trust. The LTBT is the formal mechanism to have trustees who are AI safety experts who can appoint directors to the for-profit board. As the company achieves milestones, the trust gains in power.
If you’ve ever wondered why Anthropic just seems a little bit more courageous than the other AI companies, now you know why. It’s not just Dario personally. He’s a first time founder. Most founders doing it for the first time, when the stakes get high, chicken out because they don’t have the appropriate protections to support their conviction.
The difference partly is about Dario’s character. But part of it is about the proper structure that Anthropic has surrounded him with.
Four governance dimensions (9:19)
Aakash: Your new book has this framework at the center of it. On page 91, you lay out the financial blueprint. What are the four responsibilities that each company needs to get right?
Eric: When we talk about being an incorruptible company, people want to know what that means. They hope we’re going to talk about something exciting like strategy, business model, culture, stakeholders. No. In order to understand and build an incorruptible company, you have to understand the deeper forces that act on organizations that are invisible. You can say, well, I don’t believe in gravity. But guess what? Gravity believes in you.
So when you jump up in the air, you come right back down to earth. The same thing happens to organizations. If there are cracks in the foundation, if they’re not strong enough, they get pulled down by financial gravity.
For most founders and leaders that I work with, they don’t even know what the word governance means. They’ve literally never read their corporate charter. They think it sounds like a building permit. But actually, corporate governance is the foundation, the constitutional structure that determines if an organization is strong or weak.
Our goal is to create something worth protecting. To build a trustworthy organization. Trustworthiness is the most underrated asset on the planet right now. And in order to stockpile it, we have to really get these inner dimensions right. But then if you stockpile a super valuable asset, people are going to try to steal it from you. What do you expect?
So the foundation of all governance things is compliance. Today, pretty much every board is focused only on compliance. Making sure we follow the rules, making sure no laws are broken, making sure there’s no self-dealing. That stuff’s important, but well covered elsewhere.
Purpose is the second dimension and maybe the most important. Today we live in the era of shareholder primacy, which holds that the purpose of a corporation is just to enrich its shareholders and often just its short-term shareholders. This is a pervasive legal doctrine that has spread all over the world.
The third dimension is coherence. To what degree is everything internally aligned around that purpose? For example, is it possible that the organization might be tempted to make money by betraying the mission? This happens all the time. When a private equity firm takes over your favorite restaurant, you can taste it in the food. The ownership structure of a company has a certain taste to it. That’s how pervasive this problem is.
The last dimension is what I call integrity. Integrity means both as it would with a person, the ability to make and keep a promise, but also structural integrity. To what degree can an organization be bullied or pushed off course by outside forces?
Aakash: Which of these is the hardest? Which do companies have the most trouble with?
Eric: They’re kind of from easy to hard. Most companies do okay on compliance. They may be taking an attempt at purpose, like they have a mission statement or something. I don’t think that counts, but at least it’s an effort. Such companies are not mission driven, they’re mission hopeful.
As we get further away from the things that we as leaders directly control, it gets harder. Coherence is harder because purpose is something we fundamentally choose. It’s a matter of our intention. Coherence is an operational discipline. We have to get our employees and all the other people we touch aligned.
But integrity is certainly the one people struggle the most with because that relates to boards of directors, to our relationship with investors and our relationship with the broader financial system.
Product teams and coherence (14:51)
Aakash: When it comes to doing some of this coherence work in product building, what is the role of the executive team, the product leadership, and then the actual engineering team? Who’s responsible for what parts?
Eric: Product is usually, in a product led company especially, where the rubber meets the road. A lot of leaders tell me, I believe in product quality. I’m like, do you? Is that your corporate purpose? Or just a slogan?
For a lot of people, it’s just a slogan. When push comes to shove, when the pressure comes down, they will compromise on product quality.
People hear the word purpose and think it’s some kind of ESG thing. I remember reading an article about a big public market investor criticizing Unilever. Unilever said it wanted to infuse purpose into all of its products. The hedge fund guy was like, at the point that we’re talking about the purpose of Hellman’s mayonnaise, I think we’ve lost the plot.
But actually Hellman’s mayonnaise is food. Its purpose is to nourish and delight the people that eat it. Imagine an efficiency consultant shows up and says, hey, product manager, I just realized we could save 5% of cost by using this carcinogenic ingredient. Or hey, why don’t we lace it with the tiniest bit of heroin? No one will notice, but it will make the product super addictive.
If you think the goal of the company is to enrich its shareholders, it’s very easy to be tempted. The product team is where these conflicts ultimately get born out. In engineering is often where you’ll find the people I call the torchbearers. People who don’t have the fanciest title, maybe not the highest paid, but they’re the people with the moral authority who really believe in the mission.
If you’ve ever had the job of being a torchbearer in a modern company, you live your life under siege. Every day some person is in your office with a spreadsheet asking, what is the ROI of doing the right thing? The ROI of doing the right thing is negative by definition because the costs are tangible, but the returns are intangible.
Financial gravity explained (19:22)
Aakash: Talk about forces pulling us away from the right decision. You write about this concept of financial gravity where success is actually pulling companies away from their mission. How does that work?
Eric: Financial gravity is this force I have been doing battle with my whole career. I have witnessed it in the companies I’ve helped start and I’ve faced it in the companies I started myself.
The privilege of being the Lean Startup guy is I get to help lots of people start companies. People call me all the time for advice. I’m thinking about starting a company. I want to raise a Series A. I want to prepare for an IPO.
So I’ve watched this process as companies lose what made them special. There’s this force seeping into all the decisions. It’s not necessarily conscious. People say they still value quality. When push comes to shove, they feel like it’ll be easier to compromise just this one time. Next thing you know, it’s death by a thousand paper cuts.
Jim Senegal, the founder of Costco, called it literally like taking heroin. You raise prices a little, you get away with it, your stock price goes up. Now that’s baked into the forecast. And think about how many startups have had this problem. You want to raise money, you need consistent growth, you want a hockey stick. The compromise you made in one month, you have to make again the next, or else the numbers go down.
Gravity is about the feeling you have, unconscious, that if you comply with the values of those who have more than you, you will get ahead. It’s about your imagination of future transactions.
Costco origin story (20:42)
Aakash: One of the coolest case studies you talked about was Costco, where they built this governance fortress. Can you walk us through how?
Eric: Costco is the classic exception to almost every rule. An economist was giving a lecture about how only family-run companies can sustain a mission over multiple generations. At the end they were like, yeah, also Costco for some reason. Costco’s not family run, but they seem to be the exception.
To understand Costco, you have to understand Saul Price. He’s the father of modern retail. Walmart is called Walmart because Sam Walton was paying homage to Saul Price’s company FedMart. When Saul was a lawyer, he learned about fiduciary duty to clients. When he became a retailer, he thought, who’s my client? The customer is my client.
When competitors would try to undercut him on price, he would literally put up signs in his own store saying, don’t buy this product from me, you can get it cheaper down the street.
FedMart went public. Financial gravity pushed in. Saul took it private. Found great investors. They became the 51% owner. They fired him anyway. Changed the locks on his door in 1975. By 1982, within seven years, they completely destroyed what Saul had built over more than twenty. No FedMart stores exist anywhere in the world.
Saul took two weeks off. He leased the office upstairs from FedMart. Created Price Club. One of the people who left FedMart with Saul was Jim Senegal, who started as a stock boy and worked his way up. After a few years, Senegal’s new company merged with Price Club to form Costco.
Financial gravity doesn’t like Costco any better than it liked FedMart. Every few years Wall Street makes an attempt. My favorite quote is a Wall Street analyst who wrote that Costco is spending what could have been shareholders’ profit on making a better experience for customers. That was a criticism, not a compliment.
If you had invested $10,000 in the S&P 500 in 1985, you would have $151,000 today. If you’d put that same money in the Costco IPO, you would have $8.7 million.
Saul Price understood you have to have a fiduciary hierarchy that puts shareholders last. Not because shareholders are not important, but because shareholder value is the exhaust from the engine. When you put the exhaust in the intake, you get deeply confused.
Johnson & Johnson warning (28:43)
Aakash: Can I copy paste these governance features and will I have an incorruptible company?
Eric: Absolutely not. These features make sense for Costco’s unique structure, but every company has to have the structure that makes sense for it. Just because you have a structure that resists pressure doesn’t make you incorruptible. The killer that’s about to kill you is already inside your house. Most of the time when companies collapse, the problem is internal.
In 1910, Robert Wood Johnson II went to work at his father’s company, Johnson & Johnson. He took control in 1932, in the depths of the Depression. He opened factories when others were closing them. Raised wages during the Depression. Before the company’s IPO during World War II, he created the Johnson & Johnson Credo. Patients, doctors, and nurses first. Employees second. Communities third. Shareholders last.
He had it carved into eight-foot-high limestone blocks at headquarters. While he was alive, it worked. He fired his own cousin. He fired his own son for violating the credo.
But he eventually died. Stone doesn’t enforce itself. Starting in the 80s and 90s, scandal after scandal. The worst was when they put asbestos in the baby powder and covered it up. The people who did it walked by the credo every single day. It’ll most likely be a $10 billion settlement.
A good corporate structure can protect a pre-existing ethos, but it doesn’t cause it to come into being. We need both.
Mission lock vehicles (31:25)
Aakash: You can’t just print them out like Enron. You can’t carve them into stone like Johnson & Johnson. But you can do something you talk about, which is a mission lock vehicle. What is that?
Eric: I want to warn your listeners that it’s going to sound weird. There are a lot more kinds of organizations out there than most people have been trained to encounter. Even if you’ve gone to business school, you may never have heard of this structure. Most VCs have never heard about it. We live in a corporate monoculture.
But just because this idea is new to you doesn’t make it new. The German optics company Zeiss had this structure in 1885.
Just like in political science, having only one decision-making body is intrinsically unstable. You can have checks and balances between multiple branches of government. So the mission lock vehicle is a separate entity, a set of trustees or stewards who oversee the performance and hold accountable the board of the for-profit company.
This structure is more common than you realize if you’ve ever eaten a Hershey chocolate bar, shopped at IKEA, been to Patagonia, shopped at John Lewis Partnership, had a Vanguard Mutual Fund, or taken a Novo Nordisk medication.
Companies with the industrial foundation structure are six times more likely to live to year 50. We’re talking about 10% versus 60%.
In 1920, Marie Krogh was living in Denmark. She was one of the first Danish doctors. She had contracted diabetes, which at the time was fatal. Her husband August had just received the Nobel Prize. At a lecture, she sat next to a scientist who told her about a breakthrough in Canada, the isolation of insulin.
She convinces August to extend their trip to Canada. They see the research. They ask the Canadians for a commercial license to bring the cure back to Denmark. But everyone was uncomfortable. They said, if we have a life-saving medicine, what if our future descendants feel the temptation to charge whatever they want?
They created the Nordisk Insulin Laboratory as a for-profit with a nonprofit parent. That foundation has acted as the mission guardian for more than a hundred years. The company is now Novo Nordisk, one of the largest companies in Europe, with a market cap larger at one point than the GDP of Denmark.
In the early 2000s, the trustees stopped the for-profit from doing something really stupid. It cost everyone money. It pissed everyone off. But it protected the research program that enabled the invention of GLP-1. More than $500 billion of shareholder value created by having the long-term perspective.
AI tooling and Answer.AI (40:39)
Aakash: You’re not just the Lean Startup guy. You co-founded Answer.AI with Jeremy Howard with nine people, no managers. You also built Solve It. What does a typical workday look like inside your tooling stack?
Eric: My life is so complicated. It’s different every day. The true privilege of my life is that people like Jeremy have entrusted me to be on the entrepreneurial journey with them. I’ve been a founder, I’ve been a CEO, but I’ve been a co-founder too. I like both jobs.
Jeremy Howard is a machine learning legend. For many of the people listening who got their start in machine learning as a programmer, it’s not uncommon that they learned deep learning from Jeremy on one of his fast AI courses.
When he was ready to take his research and turn it into a research lab, he was concerned. He didn’t want to wind up like everybody else. The ability to wrap him in an incorruptible structure unlocked the investment opportunity for his VCs. Jeremy is a person of tremendous integrity. If he couldn’t do it in a way that protected his principles, I don’t think he would have done it at all.
One of the products we make is called Solve It. It embodies our philosophy that AI should never be used to replace people. It should be used to augment human creativity.
Why AI writing fails (44:42)
Eric: When generative AI first started getting good, I was excited to use it for writing. But it really did not work very well. Everything you write winds up sounding like garbage. Just like private equity has a flavor, AI has a sound. It’s conforming you to the training distribution. These chatbots are mimetic conformity machines.
When people are pumped to use AI to replace humans, they’re sanding all the rough edges out of their intelligence. When someone uses the phrase AI, they are literally taking sides in an age old debate about the nature of intelligence. If you really think an LLM is intelligence, you’re siding with the postmodernists who claim that intelligence is just language.
I couldn’t use AI effectively in my writing process until we invented Solve It. It allowed human-in-the-loop AI interactions where we blend what the human and AI are doing with true shared context. You can see everything the AI can see. You can edit its responses directly. If it does something wrong, you just replace what it said.
Because LLMs are autoregressive, the more wrong answers in a chat history, the more it learns that wrong pattern. So by replacing wrong answers with correct ones, the LLM gets smarter as we go.
Solve It live demo (47:27)
Aakash: How did you write specific book chapters with your actual artistic integrity and voice?
Eric: This is Solve It. I have tons of dialogues going back months. Every single one is like a canvas. I mount my work onto it and then work on it myself. Solve It’s job is to help me, the human being, do good work, not to tell me what to do.
There are three kinds of messages. Green ones are notes, raw markdown. Orange ones are prompts, where I ask the LLM to do something. Blue are code, Python. We developed an Eric Ries writing style guide. A list of things to avoid. All the AI-ese stuff. A briefing about what the book is about.
I had more than 600 test readers using Help This Book. A CSV with more than 10,000 reader comments. My daily ritual for months was starting my day with the new comments. But at a certain point it gets overwhelming. So I wrote software to summarize the comments per chapter. Here’s what people like. Here’s problems they’ve flagged. Here’s suggestions.
The most common thing I would do is write something and then ask it to analyze it for me. Is this any good? What’s good about it? What’s not? You can’t just believe everything it says. You have to be careful because the feedback can ruin your authorial voice.
But once you figure it out, writing is the loneliest, saddest activity in the world because it’s just you and the page. It’s very nice to have a companion, even if at some level it’s kind of telling you what you want to hear.
When I figured out what I liked, I would put it back as if it was the LLM’s first suggestion. So the LLM appears to get smarter as we go. All the context bloat you see with something like Claude Code is removed. Only the information it really needs is present.
Aakash: You’ve basically created an agentic harness within Solve It with various code blocks and context blocks. A lot of people build these out as Claude Code operating systems. You built this out in Solve It in a visual way.
Structuring Answer.AI as PBC (57:39)
Aakash: A lot of PMs eventually graduate to founder. What were the choices you and Jeremy made to structure Answer.AI properly so it’s incorruptible?
Eric: Answer.AI has certain advantages. Jeremy is a legend, so investors were keen. We sat down with our investors and had a real conversation about what is the right long-term structure to protect the integrity of this research.
We converted to a Public Benefit Corporation. PBC sounds similar to B-Corp certification, but it’s something much more fundamental. PBC writes your corporate mission directly into your charter. That gives your directors permission to pursue that public benefit even if it conflicts with shareholder value.
Here is Answer.AI’s defined public benefit from our legal charter in Delaware: to develop and increase society’s understanding and awareness of the benefits and risks of artificial intelligence, by researching, teaching, and discussing the proper uses and benefits of AI, as well as developing useful and practical applications that could be ethical and beneficial.
Can you tell an engineer helped draft this? Most PBC statements are much shorter and loftier, but Jeremy and I were very worried it would be misinterpreted. We tried to be exhaustive.
Remember Silicon Valley Bank? Its mission statement was to help innovative companies move bold ideas forward. That was not their legal purpose. Their actual charter said the purpose was to engage in any lawful act or activity. Courts interpret that as maximize shareholder value. So the bank lobbied Congress for permission to make risky interest rate bets. It would have been illegal before. They got the laws changed. Then they collapsed.
The mission guardian is who has the responsibility to make sure the company stays on mission. For Answer.AI, Jeremy and I are the mission guardians. We vested voting and board control in him.
An AI research lab reflects the research agenda and vision of the founder. If you can’t trust him, you have no business investing. But when recruiting, several people said they wouldn’t have taken any other job. For the chance to do this, they’ll come work here. To recruit high caliber talent, we had to make a promise. How can I be sure this technology will be used appropriately?
Instead of saying trust the nebulous future investors, we could say trust Jeremy. That was enough for most people. As the company grows, we will expand to constitutional governance and things like the Mission Lock Vehicle. But we’re a young company. It hasn’t been necessary yet.
Aakash: So you don’t always have to do everything from the get-go. Anthropic did some major stuff at Series C.
Eric: This is really important. You can sequence it.
Build Measure Learn today (1:08:07)
Aakash: Build, Measure, Learn. I think it is the most famous business framework probably ever. You created it. Does it break for foundational AI? A $10 billion training run takes two years.
Eric: In order to answer that, you have to understand that Build, Measure, Learn is not a set of tactics. It is a general purpose statement of principle. The value in an entrepreneurial situation of high uncertainty is in the learning, the reflection on each idea. Whoever has the faster velocity through that loop has the advantage.
That’s true whether the thing takes a month or a year or ten years. I’ve used it on combined cycle power plants that take decades. On cancer therapies. I built the Long-Term Stock Exchange. It took me ten years.
What matters is not how fast you go relative to dating apps. It’s can you go faster than the convention in that industry?
In the AI foundation model wars, the labs that release more quickly, both on product and research, have a decisive advantage. The ones that put out one supermodel and go dark, customers don’t come to rely on them.
It’s not just that you learn more. It’s a matter of trustworthiness. If a company’s on fire, if they’re cooking, like Anthropic has been cooking lately, you start to feel like you should build on top of Anthropic because your app will just magically get better.
The pressure to standardize on the leading vendor is super high. Claude Code emerging as that standard is one of their huge competitive advantages, mostly powered by unparalleled product velocity.
Aakash: Is Sam Altman just a bad CEO or is he trapped in a bad structure?
Eric: Everyone always wants me to comment on Sam. Sam has been nothing but lovely to me. From my interactions with him, he’s done nothing but be high integrity and kind.
I think the focus on Sam the person obscures the structural problems that need to get fixed, not just for OpenAI, but for all of these companies. AI governance is probably as important as getting the technology right.
I would rather the focus be on what is the appropriate governance structure, who should have insight, who should have control, rather than Sam v. Elon.
I’m a believer that we should have the equivalent of a sovereign wealth fund, like the Alaska Permanent Fund. That is a common structure when you have a natural resource explosion. In exchange for all these regulations the AI companies want waived, including training on all copyrighted material without compensating authors, I’m an author. I’m in the training set.
I always tell people, if you think being in the training set is bad, try being excluded. Everyone’s in a no-win situation. It’d be smarter to have compulsory licensing. Establish the rule that they’re allowed to train on everything, but some of the profits have to go back to the people who contributed that cultural worth.
Humanity itself created this technology as a collective act. The people who did the matrix multiplication played an important role. But the raw material required the silicon chips, the RLHF, people working in low-wage countries, and all this cultural output. We should have shared governance with broadly shared prosperity.
Where to find Eric (1:15:45)
Aakash: What are all the things you have going on? Where should they go?
Eric: The most important thing is to go to incorruptible.co and buy this book. The reason you want to get it now and not wait is that if everybody waits to see if the book becomes a thing, then it doesn’t become a thing. If you actually like an author, pre-order their book. Get it right away when it first comes out.
Rather than just going on Amazon, go to incorruptible.co. We have lists of all the independent bookstores carrying the book. Walk into the store or call them. Tell them you want 10 copies on launch day, May 26.
We have implementation guides for people who want to know how to actually do it. Down to the level of the docs. We also have readers guides for different roles, whether you’re a founder, board member, employee, or consumer.
If you want to take your company public and not lose its soul, learn more about LTSC.com.
If you’re an early stage company and want legal counsel that won’t charge by the hour, tryvirtual.com. Flat fee for all the boring administrative back office.
For Lean Startup trainings, go to leanstartup.co.
The best way to follow me is to get on my mailing list at incorruptible.co or follow me on social media.