埃里克·里斯：MVP不是半成品，公司结构才决定会不会作恶

Ries notes that The Lean Startup came out in 2011, and each year he expects it to be the end of the story but new work keeps arriving. Today he is building the Long-Term Stock Exchange and advising founders and large-company executives. His years of work on corporate governance, aimed at long-termism and humanism, have recently pulled him toward top AI companies.

He says he never expected the impact. During the financial crisis he blogged anonymously because people warned it would be career suicide, and the few startup bloggers of the time wrote to him because they saw his links in their referrer logs. At his previous startup IMVU, the team shipped about 50 times a day, and VCs who asked him to bring that "magic pixie dust" to their companies sometimes yelled at him in meetings.

Ries describes his first-principles thinking as having two parts. Much of his work is descriptive, giving names such as pivot to things startups already do, because without a name you cannot reason about them. The prescriptive advice then follows by deduction from the hypothesis that a startup is an experiment, and the concepts came from a great deal of trial and error rather than years alone in a cave.

Ries expected to lead a movement into battle, but he says no opposing army ever charged back. The idea went from controversial to obvious, and some early opponents later claimed they had always agreed. He decided that the real victory is the stories people tell him about how the book helped them, and that because it is a scientific theory, whatever works counts as Lean Startup.

He says the common misconceptions have not changed in a decade: that lean means cheap, that it opposes vision, that experimentation leads to local maxima, and that a pivot is prescriptive. He once went back to the book to check and found the vision point addressed on page nine. He is fine with criticism but tired of reflexive repetition of the same old claims.

Asked what he would change, Ries says everything, which is why he refuses to write a second edition. He cringes at the scaling chapter, which he says made enterprise adoption sound easy, and he later wrote a full book on it. He also regrets not saying more clearly that entrepreneurs should change the world "for the better," a line he found missing when he reopened the introduction.

Responding to a question from Linear founder Karri Saarinen about rising user expectations, Ries says an MVP is not a tactic but the most efficient way to validate a hypothesis. Expectations are often higher in industries like data-center equipment or jet engines, yet even there the cost of offering something people don't want is a polite no. He advises finding 10 customers rather than 10 million, and letting them tell you what is awful.

He distinguishes a product launch from a marketing launch, since you can put a product in front of customers without announcing it. He adds that an MVP is flexible: if the quality bar is high, do the one critical feature at that quality and then add more. In pivot stories, he says, often 90% of the first version's work is thrown away however polished, and quality is defined by customers you may not have identified yet.

To show that less can be more, Ries describes IMVU's avatars, which were stuck in chairs because the team lacked inverse kinematics for walking. Customers complained, so he hacked in a crude version where the avatar simply vanished and reappeared elsewhere. Customers soon rated IMVU as more advanced than The Sims because, as one explained, you teleport instead of waiting for a character to walk.

He gives a second example from an equipment maker with PhDs planning a three-year, $18 million development. Ries convinced them to take a brochure to customers first, and customers laughed them out of the room because they did not pay for efficiency, only for building data centers. The lesson is that these insights seem obvious in retrospect but happen to everyone.

Replying to Ben Silbermann's remark that you don't know what minimum or viable means, Ries says criticisms of MVPs are really frustration with how uncertain startups are. His first tip is to list the features you think are necessary, cut that in half, and cut it in half again. The MVP is a U-shaped curve where being near the optimum is fine, and people are usually off by one or two orders of magnitude.

If a team struggles to define an MVP, he says the problem is usually upstream: they are unclear about what they want to learn, often because they do not want to admit they don't know. Ries tells how the Long-Term Stock Exchange spent two years on a partnership agreement and an SEC approval process, and repeatedly told his team the goal was learning, not just getting a contract signed.

At the end, the SEC withdrew its approval and the partnership collapsed, which he calls one of the worst things that ever happened to him and later the best, since it led to the pivot that got the exchange approved. All of the contract work turned out to be wasted, but the people who lived through it moved much faster afterward. His lesson is that if you cannot admit something failed, you cannot learn, and an MVP is meant to reveal fatal problems while there is still time to act.

Lenny notes that many beautifully crafted startups are failing, and asks when to invest in craft. Ries says he finds beauty in simplicity and would rather see one thing done extraordinarily well than twenty done halfway. He points out that perfection only makes sense relative to a customer, so you must know who the product is perfect for.

He supports craft in how you work, such as good code, the right abstractions and strong people, because that keeps options open and increases speed. If someone wants to build for themselves or has a core value they would rather go out of business than violate, he respects that, as long as they do not pretend to test it. Where people get into trouble is using craft as a way to avoid feedback.

Ries adds that people who lack taste can acquire it, because experimentation will teach the habits of mind and patterns that work. Craft as a signal to customers is also part of a promise, so an MVP that lacks it would not test that thesis. In his view the supposed trade-off between craft and MVPs is a dichotomy that does not really exist.

Ries urges people to be willing to get fired for doing things the way they believe is right, saying it was a career accelerator for him because the right people are drawn to you. He then argues that a startup failing is not even in the top ten worst outcomes. Worse is being stuck in a "zombie" company you hate, or selling a company that becomes something you find abhorrent, and he says founders face a mental health crisis that is not discussed enough.

For founders stuck in such companies, he says the hockey stick may never come, and that it is always darkest before the dawn but also dark when you are dead. He asks them to introspect: if you could magically start a new company today, would it be this one? If not, he says, it is okay to move on, and the discussion becomes tactical, such as shutting down, returning money or pivoting hard.

He ties the pain to ego identification, since many founders feel that if the company dies, they die. He hopes founders can learn to disidentify from the company so it can end without ending them. The hardest step, he says, is admitting the facts as they are, and he criticizes investor updates that say "we're crushing it" right before a shutdown notice.

Lenny cites research that about a fifth of consumer startups and about 40% of B2B startups pivot to a completely different product, and Ries agrees the numbers sound right. He used to tell the IMVU story two ways, as an idiotic series of pivots and as a series of genius moves, and says both are true. Because founders often present sanitized stories, he says it is hard to learn the truth from the outside.

Ries defines a pivot as a change in strategy without a change in vision, but says the vision itself evolves because real life forces trade-offs. He tells of finding an old whiteboard in his own handwriting from a founding meeting, listing ideas he thought were his co-founders' bad ones. He had no memory of changing his mind, which he uses to argue why science writes hypotheses down: human psychology is buggy.

Lenny lists pivots such as Slack starting as the game Glitch, and Ries says there is usually a thread from the first thing to the second that makes sense from the inside. He notes that even Facebook's founders, while it was already succeeding, were working on other projects, and that being a visionary is more about courage to dance with the world than knowing the future.

Ries says if you are asking whether to pivot, you probably already know the answer. Likewise, anyone asking him whether they have product-market fit does not have it, because when it is working there is no time for such questions. The first step is to admit the truth to yourself.

His practical suggestion is to give yourself a fixed period, such as six weeks or even one weekend, to take decisive action. If you cannot agree that it is not working, spend that time going heads-down on the one thing that matters. If you do agree, have someone dedicated full-time to a new idea on a deadline, and have everyone say what they would do if they could start the company over, which sometimes reveals that everyone secretly wants the same thing.

He compares stuck startups to Thomas Kuhn's paradigms: nothing productive is left to test, and incremental metric gains such as 10 to 10.5 percent are not what is needed. If several ideas are equally appealing, run them in order, each for a month. If the team is exhausted, he says it is fine to return the money, and sometimes a team needs to fix the capital structure to try again.

On AI, Ries says people are not really afraid of AI but of corporations with AI. He quotes the idea that software reflects the organization that produces it, and says that will be even truer for AI, where companies transmit their values. He argues tech has underinvested in governance structures that embody values, which is why many people feel betrayed by companies they helped build.

He calls AI at bottom a management technology. When he tried AI agents, he found they never say no, so they would spawn subagents that pass the buck in an infinite regress, such as when he asked one to write a 30-day course and it kept delegating the task. He also expects AI to change management by letting leaders summarize an entire company without layers of middle managers.

Many industry people are having an existential crisis about what AI work will have a positive effect, and Ries offers a framework for ethical action under high uncertainty. Since no one knows which future will arrive, choose actions that make ethical sense across many scenarios, such as more transparency, state capacity, competent leaders and companies committed to values. That, he says, moves people from paralysis to action.

Ries shares a story of someone using GPT-4 to send thousands of personalized cold emails to Shopify stores, and sees it as experimentation at a new scale. He worries communication channels will flood with AI-generated messages, but is optimistic that AI agents could make marketplaces fairer. A buyer's agent could filter proposals against a procurement policy and surface only credible ones, without clever positioning or personal connections.

Ries says founders today face far more questions about trust, because their technology is more invasive and powerful. He argues that companies claiming to change the world cannot also disclaim moral responsibility, and describes working on companies straight out of science fiction, from San Francisco-to-Tokyo travel in an hour to weather control to biological AI. Asked why anyone should trust them with such power, founders typically answer that they are good people.

He says that answer fails because no individual can make promises on behalf of an organization, since founders can be replaced or investors can swap management. In his view organizations are living superorganisms that founders nurture but do not own, and promises must be embodied in the structure so they hold when the founder is gone. He cites medical ethics and rules requiring doctors to run care organizations as models.

He tells of asking an ex-Googler how likely the company is to file its quarterly report on time versus avoid being complicit in killing someone for profit. The first was certain and the second was not, which Ries says shows that companies build apparatus for what they care about. His conclusion is that profit should mean maximizing human flourishing, not extraction, and he says ESG scores such as Philip Morris's show the current approach is broken.

Ries argues that companies built around human flourishing can be a competitive advantage, and he describes one that tells staff to treat customers like their own parents and enjoys remarkable loyalty. He says builders focus on product and too little on the organization, and calls for a new movement to build such companies.

Asked how founders can start, Ries says resources are still thin, but he lists options: becoming a public benefit corporation or social purpose corporation, placing a share of the company in a foundation with stakeholder trustees, a board mission pledge, and a financing vehicle called an LTSPV whose investors commit to support the mission. His key warning is that it is always too early until it is too late, and founders who postpone often find the IPO process too far along.

He suggests asking your lawyer a hypothetical: if Philip Morris offered a dollar more per share to buy your company to sell cigarettes to children, would you have a fiduciary duty to say yes or no? He says with standard Silicon Valley incorporation documents many governance experts would say yes. Ask for pros and cons of each protective option and then pick a few to start with, since there is no one weird trick.

Ries says this is an open secret rather than a hypothetical. Toyota has unusual governance, foundation-controlled companies such as Hershey exist, and in some European countries 20 to 25% of public companies are structured that way. He also cites OpenAI's nonprofit-with-subsidiary structure and says a study of Danish foundation-controlled companies found they outperformed on financial performance, which suggests shareholder primacy is wrong even for shareholders.

He calls his idea a spiritual holding company that is responsible for the soul of an enterprise. He says it is also a practical recruiting tool, such as when tenured academics leaving labs want assurance the technology will not be misused. He urges founders not to take it on faith but to think for themselves and try what makes sense.

In the lightning round, a discussion of a well-made humidifier leads Lenny to observe that great products often come from one person's control. Ries describes Toyota's chief engineer, who has moral authority over a whole car line with no direct reports and makes the trade-off decisions that keep the design coherent. He wishes more founders knew about this role.

Asked which of his concepts did not catch on, Ries picks engagement loops, the counterpart to viral loops. He argues customers come back only through synthetic notifications, organic notifications such as PayPal's "you've got money," or positioning that makes the product part of their life. He thinks these can be studied with the same precision as viral loops, and Lenny notes that Ries was also the first to describe growth as loops.

Ries closes by asking listeners to try his ideas and tell him what happened, including if it did not work. He then argues that founders collectively supply the new companies that the current financialized system needs to survive, so they can set the terms and insist on shared values. He hopes they will defend those values more ardently.