Mental Models That Change How You Think | Bill Gurley
What this covers
Bill Gurley spent years on Wall Street, built his career as a partner at Benchmark, worked through Uber’s hypergrowth era, and now serves on the board of the Santa Fe Institute, where he studies complexity and systems thinking.
In this episode, Bill shares the mental models he returns to most, including systems thinking, second- and third-order effects, and the importance of understanding both the bedrock of your field and the bleeding edge.
He explains what separates great founders, why storytelling and product instincts matter, how he uses AI across different models, and what he sees coming in open source, China, stablecoins, tokenization, payments, and venture capital.
Timestamps: 00:00 Key Mental Models 02:02 Investing Journey and Key Players 05:21 Knowing the Bedrock of the Industry 08:50 Obsessive Learning in Founders 10:04 The Silent Edge 11:44 Surprising AI Use 13:13 The Future of AI Models 14:17 Global AI Regulation 18:12 Impacts of AI on Investing 19:53 Are There Limitations on Training AI Models? 23:04 Would You Sit in the Back Seat While Your Tesla Drives? 24:15 Non-Consensus Opinions 24:53 Are We Overfunding this Buildout? 29:40 The Role of Retail Investors and Tokenization 34:26 What is a Stablecoin? 37:58 Competitive Mode: Visa and Mastercard 39:55 AI and Debt Analysis 45:05 The Craft of Storytelling and Writing 48:07 Founder Advantage: Product Instinct 50:12 Real World Lessons from Working With Uber 52:10 Inside Benchmark’s Success 59:42 What is Success for You?
Follow Bill Gurley LinkedIn: https://www.linkedin.com/in/billgurley/ X: https://x.com/bgurley?lang=en Check out Runnin’ Down a Dream: How to Thrive in a Career You Actually Love - https://amzn.to/4dP6VZ1 ------ Thank you to the sponsors for this episode:
+CoinShares: Delivering Reason to Digital Asset Investing. https://coinshares.com/
+Granola AI, The AI notepad for people in back-to-back meetings: https://www.granola.ai/shane
Check out the Granola Notes: https://notes.granola.ai/t/c8d46a3e-a682-41ed-a649-933df79f6cfe-009c2hma
+HeyGen is a message-first AI video platform that helps people and AI agents turn ideas into professional video in minutes. Try for free at https://www.heygen.com/tkp
+LMNT: My go-to zero sugar electrolytes — get a free LMNT Sample Pack here: DrinkLMNT.com/TKP
*Shane Parrish* Instagram: https://www.instagram.com/farnamstreet/ X: https://x.com/ShaneParrish LinkedIn: https://www.linkedin.com/in/shane-parrish-050a2183/ Books: https://fs.blog/books/ Website: https://fs.blog/ Newsletter: http://fs.blog/newsletter
*The Knowledge Project* is a show featuring in-depth conversations with the top CEOs, investors, and business leaders to uncover the timeless principles that drive success. Learn more at https://fs.blog/podcast
Source description (no synthesized summary yet).
Understanding foundational knowledge across domains—from field history to cutting-edge developments—combined with systems thinking and obsessive learning creates unfair advantages for founders and investors navigating complex, nonlinear environments.
- Mastery of field history (Picasso as realist, chess history, animation classics) provides differentiating context that most competitors lack
- Systems thinking reveals second and third-order consequences that linear thinking misses, preventing costly downstream errors
- The combination of deep historical knowledge plus obsessive edge-learning positions individuals to exploit disruption cycles before others recognize them
This asset isn't compiled yet
You're seeing its claims, ranked. Compile it to build the argument threads, weight them, and check each claim against your library — the full view.
Understanding what Wall Street values in companies allows venture capitalists to think backward from eventual exit outcomes, making decisions about early-stage companies by considering how they will appear to public market buyers.
“I've always thought of Wall Street as the buyer of the product that venture capitalists create because of the eventual liquidity is either an M&A or an IPO. And now the price is being set by that group and that institution. So if I know what they value, even if we're starting at a very early place, two people in a PowerPoint, you're still thinking about when this thing grows up, is it going to be something they're excited about?”
An agricultural metaphor illustrates the systems advantage of open-source innovation: imagine two agricultural societies where farmers come to market; one society just sells goods to each other, while the other society forces farmers to share best practices with all other farmers. The second society will evolve faster due to knowledge sharing.
“I'll use a simple metaphor but imagine you have two societies and both agricultural societies and one of them when all the farmers come to market they just sell each other goods and then they go back and the other society when the farmers come to market they're forced to share best practices with all the other farmers. Which which one of those is going to going to evolve faster?”
Complex systems are multivariable nonlinear systems that can behave one way for a long time and then one variable can switch causing them to behave another way entirely, making linear prediction impossible and requiring awareness of cascading second and third derivative consequences.
“I would describe complex systems as multivariable nonlinear systems. And multivariable nonlinear systems are very hard to predict. They can behave one way for a long time and then one variable can switch and they can behave another way. The weather, stock markets, all these things. There's consequences that can be first, second, third derivative.”
Value investing's core principle—that an asset is underpriced relative to its future worth—can be extended to high-growth companies like Amazon if you believe in network effects and their ability to sustain unreasonable growth rates for very long periods.
“He would say that, you know, value just means that the asset is underpriced relative to what you think it will be worth in the future. I spent a lot of time talking with Bill about network effects. And if you believe in that, then Amazon might be able to grow at a unreasonable growth rate for a very long period of time, which he believed.”
In capital-abundant markets, founders choose venture firms based on reputation (network effects from past success), founder understanding of their specific domain, and genuine excitement about the business.
“if you're successful as a venture capitalist, people want to work with you... but their stamp of approval of you will carry weight in and of itself. And so there some people have said it's the only investing category where there are network effects because once you have a reputation it it you have an unfair advantage in deal flow.”
The venture capital industry structurally bends toward youth because it's a high-hustle business where newer VCs have more available time to obsessively study emerging technologies than older VCs with families and responsibilities.
“I think the the whole industry bends towards youth for that reason and because it's a hustle business like there there's there's always a rock you haven't looked under and you know age brings children and homes and and and other requirements you get tied to and responsibilities and you're just not able to go spend 80 hours a week studying YouTube”
When mobile phones emerged, no engineers had written mobile apps before, so entrepreneurs who obsessively learned on that edge early figured out the implications first and gained massive advantages.
“when the mobile phone came out, there were no engineers that had written apps for mobile phones. and a few people got on that edge and figured out what that meant. And that requires obsessive learning on the edge.”
The venture capital community's belief in power laws and increasing returns has made investors systematically more willing to invest on the come and take on larger risks, expecting that winner-take-most dynamics will eventually justify massive early losses.
“the investor community writ large has slowly become aware of and believes it strongly in increasing returns and power laws. And so over time, if they all believe that, they're going to be more willing to invest on the come and take risk.”
Tokenization of private company equity creates risk of speculation and manipulation if it occurs on assets without proper financial disclosure regulations, which could harm both speculators and the company's operations.
“there's a there's a risk with tokenization, especially if it happens on assets that don't have regulation around financial disclosure that you get a ton of speculation and even worse manipulation.”
Companies like Stripe intentionally remain private to avoid the volatility of public market pricing, allowing management to control valuations through occasional negotiated transactions rather than continuous market-driven pricing.
“One of the reasons they're staying private is so you don't have that dynamic because they have more control over sort of like the market cap pricing”
Most of the developed world established instant bank-to-bank transfer systems 15-20 years ago (UK faster payments, Argentina's Picks), but the US government has been prevented from implementing similar systems by financial industry regulatory capture.
“most of the rest of the developed world, the governments um established a an ability to do instant transfer from bank account to bank account and from bank account to a partner or retailer, whatever. UK faster payments did this 20 years ago. Recently, Argentina did it with picks in the past six years and it quickly became 60 70% of transactions and precisely because of regulatory capture the banks have kept our government from doing that.”
Moody's competitive advantage comes not from superior analysis but from being an entrenched standard that everyone trusts, making them difficult to displace even if AI could perform their analytical function better.
“I think Moody's power comes from the fact that um that it's a standard and everybody trusted as a standard”
Index funds' lack of time to evaluate individual shareholder votes has created dependence on advisory services like ISS that score companies with black-box methodologies not disclosed to the public, creating a problematic two-sided market dynamic.
“the index funds don't have the time to truly evaluate the what the vote should be in these situations. And so the they rely on these services, but these services have been built. They play this game that that that it's not particularly um settling, but they they score you, but they score you with a black box. They don't tell you how they score you, right?”
ISS and similar proxy advisors started as fraud prevention mechanisms focused on corporate governance risk, not shareholder value, causing them to oppose creative compensation structures that align founder incentives with stock performance.
“They started from a place of corporate governance where they were looking out for fraud. And so so risk mitigation rather than shareholder interest.”
Passive index investors currently own such large percentages of public companies that active investors' voting power is diluted, and would only be restored if passive investors abstained from voting.
“this wouldn't be a problem were it not for because it's the large number of shares held by the uh the passive it one thing that would be really great is if they just wouldn't vote um because then the people that are active shareholders would have more of a say”
Public equity investors marking positions to major indices forced many active managers to practice 'closet indexing'—maintaining similar holdings to indices to avoid underperformance—reducing genuine active management capacity.
“the public investors got really scared because they were marked to the index and they ended up doing what people call closet indexing to make sure that they didn't lose out. And like when the MAG 7 took off, um if you didn't own those, like you had a bad year”
Benchmark's founding structure as an equal partnership with no CEO or hierarchy was designed to address observed problems in traditional hierarchical venture partnerships where senior partners kept disproportionate economics.
“the founders of Benchmark had been at hierarchical firms where they felt like the senior patriarchs were maybe taking too much of the money and too much of the credit and not doing the work that was imperative for the firm's success.”
Equal partnership eliminates competitive zero-sum dynamics where partners hoard information and avoid sharing ideas because a colleague's success might reduce their relative status.
“If one of my companies needs a new CFO and they know of one, they'll probably just give it to me right away. My company succeeding is no different than their company succeeding. So, you just create a a different dynamic.”
Beating the S&P 500 is very hard; many active investors, including venture funds, fail to outperform, with some noting that the Nasdaq-100 (QQQ) has outperformed 80-90% of venture funds.
“the buy side is a very hard job like to to beat the S&P. Some people have even highlighted the fact that, you know, QQQ has probably outperformed 80 or 90% of venture funds.”
Writing out ideas in long form—like Amazon's 6-page memo practice—forces deeper thinking about problems, reveals loose ends, and creates cognitive discipline that verbal discussion alone cannot achieve.
“that's exactly why Bezos has his six-page letter concept at Amazon. He he believes that if you have to write it out and make it stand alone and be cogent that you'll think through more of the the problems and you'll it'll be more cohesive and it'll um you you'll figure out the loose ends and you'll tie them up.”
Understanding the craft of storytelling and long-form narrative is essential for founders because they are constantly selling—recruiting employees, raising capital, closing customers and partnerships.
“storytelling is so important because in the founder case, you're recruiting employees, you're you're recruiting executives, you're raising money, you're closing customers, you're closing partnerships, you're selling all the damn time.”
Public demonstrations of expertise in writing about business and investing—like hosting a blog or publishing essays—act as a filtering mechanism that attracts relevant deal flow and founder interest without requiring active business development.
“in addition to that, in the venture world, for the founder that doesn't know you, when they see your knowledge on a subject or they see what you're talking about in their own business, they reach out to you. So, it becomes a calling card.”
ICO (Initial Coin Offering) mechanisms use auction-based allocation that matches supply and demand anonymously, similar to optimal IPO design, suggesting tokenization could improve IPO fairness if regulators allowed it.
“with tokenization, no one would invent this thing where you you cherrypick your best customers and give them this sweetheart price. No one would do that.”
Wall Street has resisted direct listings (which use auction mechanisms) because they generate lower fees than traditional IPOs; because bankers control regulatory and political channels ('greedy power grab'), they've returned to controlled oligopoly pricing rather than embracing more transparent mechanisms.
“we pushed direct listings for a while, which which uses this auction mechanism, and they could have embraced that, but they didn't. They've gone back to this kind of controlled igopoly. I I think that is an area where where tokenization like just merely getting to the first base of how the share should be allocated um could be very disruptive.”
Despite widespread passive indexing, some argue that the reduction in active investors might have increased edge opportunities for genuinely active investors willing to do differentiated research.
“But they were they had kind of reached a point where they think the number of active investors is so few that the ability to get an edge has maybe increased as a result of the of the massive indexing.”
Equal partnerships are difficult to scale because there is no CEO to make infrastructure decisions, causing coordination problems around new initiatives like maintaining a company website.
“There is there is one huge negative. So I don't want to just say it's all um the it's almost impossible to have uh because you don't have a CEO. It's hard to scale out and it's hard to have new initiatives like because there's no like oh maybe we should you know the website was always a funny one like who's going to own the website?”
Large language models have inherent architectural limitations because they are language-based and cannot natively represent mathematical and numerical relationships with the same precision they use for text.
“the next version of AI is not it's not LLM it's it's outside of LLMs it's broader than LLMs and that that we're going to run into a asymptote with these because they're language based and there's just a limit to what language what you can capture with language which is part of why they're not specifically great with math and numbers, right?”
The cumulative losses that leading AI companies are expected to incur before reaching profitability will be larger than the losses of previous generation platforms (Amazon ~$2-3B, Uber ~$15B), setting new precedents for capital-intensive ventures.
“someone forwarded me a chart this morning of the losses of the leading company in the field prior to going cash flow positive. And you look at, you know, what for for for Amazon it was like two or three billion. For Uber it was like, you know, 15 billion. And now for these companies, it's going to be way bigger than that.”
Burn rate—the speed at which a company spends capital—historically served as a primary risk measure, but modern venture companies burning $100+ million per month make traditional unit economics analysis nearly impossible.
“I always thought of burn rate as a measure of risk. 10 years ago, like it was super risky to burn a million a month. You know, today these companies are burning five billion a year. Like like you know, you're you're burning a hundred million a month or more. Like it's really hard and this may go back to financial bedrock and whatnot. It's really hard to know what your unit economics are when you're being that aggressive financially.”
The market for capital is not the constraint in venture investing; instead, pricing power and valuation disputes between different investor types are the likely locus of change if funding dynamics shift.
“there is zero lack of fund availability right now. So that's not the bottleneck. Yeah, there's the pricing would change, right? There's no con.”
Stable coins like USDC, backed 1:1 by US Treasury holdings and transferred on blockchain rails, enable instant global money transfers for pennies, directly competing with credit card systems that charge 2-3% fees.
“It's a cryptocurrency that if the company's following the regulation, I believe that USDC is is in fact doing that. Um where they have they have created a dollar fordoll holding in treasuries, US treasuries. Okay. For each each stable coin that's represented.”
Retail investors have driven outsized valuations for certain public companies (e.g., GameStop) and their preferences have created valuation gaps that institutional investors struggle to understand, suggesting retail money influences pricing.
“this has kind of played out in the public markets. I mean, I think you look at obviously like stocks like GameStop, but I I think most people believe Palunteer is a stock that retail investors really love and take it to a valuation that it's very hard for institutional investors to get their head around.”
The traditional IPO process where underwriters choose the price and select shareholders is structurally unfair to companies and wasteful compared to market-based auction mechanisms that would allocate shares to highest bidders.
“I think it is insanely unfair to the companies the way they're forced to go through this process where the bankers pick the price and pick the shareholders. There's just no need to do that. If you took a freshman computer science student and a freshman finance student and said, you know, imagine how a company should go public. They would match supply and demand anonymously like you would in any auction.”
Elon Musk's Tesla compensation package, which provides no payment unless stock rises significantly but enormous wealth if it does, should be attractive to shareholders as it maximally aligns founder and investor interests.
“The Tesla case is a great example that that package that type of package that they did for Elon. I've said this, you know, publicly. I would agree to that type of package for every company I've ever worked with, and most CEOs wouldn't take it. Uh, it basically says you don't make money unless the stock goes way up.”
AlphaGo and Tesla's Full Self-Driving are both constrained AI systems trained to specific bounded domains, not general-purpose AI systems, making their success at specialized tasks distinct from general intelligence.
“And by the way, those AI models aren't LLM based like AlphaGo is not LM based. It's a AI model trained to a very specific constraint system. And even FSD is you know at Tesla is a constrained environment like there's the inputs are the brake and the steering wheel and the gas pedal”
The trajectory and direction a company is heading matters more than the starting point, because the ultimate value and outcome is determined by where it ends up, not where it began.
“The trajectory matters more than the starting place, I think. Yeah. That that's where you're going to end. That's the out that's the output at the end of the day.”
Picasso was a wildly successful realist painter by age 14, which is visible in Barcelona Museum, but this fact is not intuitive from observing his later cubist paintings.
“Picasso was a wildly successful realist painter by the time he was 14. If you go to the Barcelona Museum, you can see that. And I don't think anyone that looks at his cubist paintings would intuit that that was true.”
China's government decision to make money transfer easy led to Alibaba and Tencent building extremely efficient digital payment systems (WeChat Pay, AliPay) that dominate daily transactions, demonstrating how infrastructure policy drives innovation.
“they they've innovated their entire payment system way further than we have because because of this the decision by the government to make money transfer easy and that means like no three-day settlement sort of exactly”
Having spent significant time in China over the past 20 years, the prevailing vilification of China in Washington and Silicon Valley seems reflexive rather than justified, particularly given the US represents only 3-5% of global population.
“Having spent ton of time in China over the past 20 years, it's hard for me to adopt this mindset of vilification that's heavy amongst many in Washington and now many in Silicon Valley. The US is like 3, four, 5% of the global population.”
Traditional partnerships in law and accounting firms are hierarchical, with senior partners taking more power and economics while junior partners must work their way up over long periods.
“most partnerships, you know, you think law partnerships or accounting firm partnerships work in a way where the the senior people have more power and take more of the economics and the junior people have to work their way up over a long period of time.”
Circular deals, where cloud service providers extend credit to AI companies to spend on the providers' own services, artificially inflate growth numbers by creating money out of thin air that wouldn't otherwise be spent.
“imagine you're a cloud service provider and you notice that this company Anthropic wants to develop this model. It's going to cost maybe $5 billion, but they don't have that money. So, you give them that money so that they can spend it. And I'm like, well, if you didn't give it to them, they wouldn't spend it. And so like the growth of everything is enhanced by the fact that you're giving money to companies to spend back on your services they wouldn't have otherwise.”
Imagine interviewing for a job at P&G or Pepsi with 20 other candidates; if you're the one person who understands the masters of marketing better than the others and can bring that knowledge into the interview, you create wildly differentiating contrast with your peers.
“Imagine you're interviewing for a job at PNG or Pepsi out of college and there's 20 people there and you're the one that understands the masters of marketing more than the others and you're able to bring that up in the interview. Isn't that wildly differentiating?”
AlphaGo's discovery of move 37 in a famous game demonstrates that AI can innovate beyond training data by searching through a constrained game space that humans cannot, but this capability may not transfer to unconstrained real-world systems.
“there's a people point to this famous game that Google Alph Go where Google implemented and the bot eventually came up with a move that was shocking to all humans... That is proof that they can innovate, you know, beyond what they're taught.”
There is zero reason why payment processing should cost 2-3%; these high fees persist only because of regulatory barriers that prevent competition.
“There's zero reason why it should cost 2 or 3%. Just zero.”
The phrase 'American exceptionalism' likely produces negative reactions from the other 95% of the planet, suggesting a blind spot in how US leaders and tech figures communicate globally.
“When people utter that word, I always wonder like imagine the other 95% of the planet thinks when they hear someone say that.”
Visa and Mastercard are duopolies created by and owned by banks; they maintain operating margins of approximately 60%—among the highest in business—because of regulatory protection and incumbent control.
“I think they will be heavily threatened by this. And historically what they've done in and by the way those two companies have two of the highest operating margins in the history of business. They have like 60% operating margins and they're they're duopies and and they were created by the banks. So and the banks have a stake in it.”
Regulation that is expensive and difficult could lead to market oligopoly in AI, and some major AI players may actually be asking for and welcoming regulation because it raises the bar against competitors, especially Chinese open-source models.
“I think the thing you know that could cut against that if the regulation gets extremely difficult and mundane and expensive that could actually lead to more igopoly and I think some of the players know that and are begging for regulation. [clears throat and laughter] Oh cuz they want that because it's a protective mode pulls up the bar against especially against the Chinese open source models.”
Websites with high cognitive load (complex navigation, mission statements, excessive information) waste user attention; simple splash pages with just essential information reflect design confidence and respect user time.
“Well, you know, it's interesting you say that because I find a lot of websites have such a high cognitive load to use a splash page with like, you know, four or five sentences or Berway's website. I totally get it. I don't there's not a lot of cognitive like [laughter] if I you know I heard this example um from a guy a couple weeks ago. He's like if I'm going to buy a sweater I don't want to know your mission statement.”
For incumbent companies, embracing new edges creates the innovator's dilemma because doing so might require admitting previous strategic decisions were wrong and giving up past commitments.
“if I'm an incumbent, it's much harder to dive into that because it might mean, it's the innovator's dilemma in a way, but it might mean giving up a previous decision I've made or saying that I've been wrong and going backwards.”
Legal AI startups invest significant time ensuring they ingest all relevant case law and understand domain-specific processes and principles before implementation, suggesting that specialized domain knowledge cannot simply be swapped out for a general-purpose model as companies scale.
“there are workflows and data modes that if you get and and also just understanding like there's three or four legal startups in the AI space, they're just spending so much more time making sure they ingest all the case law and and really understand, you know, the processes and principles there and and then you implement with them and they're writing stuff on your behalf and you're building new databases out of there. I just don't know that you then switch that to chat GBT as they climb up the stack.”
Understanding the historical bedrock and foundational figures in your field provides differentiating value in professional contexts because most people skim information rather than studying deeply, so mastery of field history stands out.
“We do live in a world where people skim. They want the gist of things. They want, 'Give me the summary. Give me the executive summary.' And I think more people would benefit by studying the history of whatever field they're in.”
China has approximately 10 open-source AI models competing intensely with each other, creating a system where all models learn from one another and can train or test each other, which enables far faster innovation than the more concentrated competitive system in the West.
“they have like 10 open- source models and so you have a situation where the competitive dynamic in China is more intense because it's more intense. Everyone's chosen to go open source and that creates a system that in my mind is capable of innovating far faster than the competitive system we have here. All the models learn from one another. You can actually have a model train another model or test another model.”
A dating site tested a hypothesis that longer user profiles would increase engagement, found it was true in the short term, rolled out the change, but discovered many months later that it actually decreased conversion—because when people knew more about potential matches at that level of detail, they were less likely to convert into actual dates; this is an example of a second derivative effect.
“I was talking to a guy that worked at one of the large dating sites. They had this idea making the profile longer would lead to more engagement. Simple, you know, heuristic. And they tested it and it was true. And so they rolled it out. They found out many many months later that it let it it was negative for conversion like when people knew more at that level.”
Most of Gurley's successful venture investments have been in marketplace companies; the lack of an initial knowledge base and codified marketplace theory required building the framework explicitly and writing it down, which became a critical asset for founder recruiting.
“If I was thinking through a problem about a new most of my most successful investments fall in this category people call marketplaces and before there was a first marketplace like there wasn't a knowledge base and you know we crafted that along the way and codified it and wrote it down”
Some people argue that if AI becomes superintelligent, it can make itself incrementally better, creating a nonlinear self-improvement curve; this is a hypothesis some people have made.
“I mean the I guess the theory correct me if I'm wrong is like the minute that they are super intelligent they can effectively make themselves a little bit better and at that point you just you enter a nonlinear curve. That that's an argument that some people have made. I I don't know that I believe it but but”
Equal partnership makes it easy to recruit exceptional talent from hierarchical firms because people can immediately join as equals rather than waiting for eventual promotion.
“The first thing is it makes it very easy to recruit exceptional talent from other firms because they're not in that situation and you immediately and I was at a firm that was hierarchical and you know even if you went back and said well I'm going to leave to go to this equal partnership and they said oh we'll make you equal well you did it because I was leaving not because it works that way”
When using AI language models, most users underestimate their capabilities and only ask single-layer queries when they could structure multi-stage prompts that handle ranking, comparative analysis, and synthesis within a single request.
“You often underestimate how much it can do. So you might ask it um to f identify the top 10 of something and then you're going to take those 10 and go study them. But you can say identify the top 10, list their pros and cons, and then rank order them based on this dimension. and then rank order them again based on another like stuff you would have done later you can just build into the prompt”
Western startups are quietly and widely using open-source Chinese AI models throughout Silicon Valley, despite this not being reported prominently in major publications.
“I would say it's kind of a quiet secret just because I haven't read it on the front page of of the journal that you know especially from a breath standpoint like a volume companies are using these models all over Silicon Valley.”
There may be limitations to current AI training approaches where models are trained on all internet data; Elon Musk represents an alternative approach where data is filtered first to remove clear untruths before using it as training data, starting from a cleaner foundation.
“Do you think there's limitations to how we're training the models now, which is sort of they're trained on all the data from the internet, including like, you know, Elon has the opposite approach where he's like, we're going to take all the data and then we're going to filter out clear untruths. We're going to use that as the starting point”
Stablecoins are a workaround for US regulatory capture; they function as an alternative payment system when the government could theoretically solve the problem with Fed Now but won't due to banking industry pressure.
“And credit cards which cost 2 and a half% but this there's no reason that it should and once again these other countries which include UK, Australia, India, China, Argentina, they've all done this but we never did it and probably won't like I at this point I think stable coins will get there faster than than the government will will be able to do it.”
A common trait among successful founders is obsessive learning driven by the need to stay at the cutting edge of dynamic technologies disrupting their markets.
“a more common trait that's related in the entrepreneurial world is obsessive learning like constant learning because the disruptions that allow for the technology waves that allow for companies to be disruptive and take market share from an incumbent are all tied to something dynamic that's happening on the edge.”
John Lasseter, the creative genius behind Pixar, tied each course of a 10-course meal to a classic cartoon he believed was foundational to understanding animation, demonstrating deep mastery of field history.
“And one year, he bought a dinner with John Lasseter, the creative genius behind Pixar. And we go to John's house and he serves us in his movie studio. He serves us in his viewing room a 10 course meal. And each piece of the meal is tied to a classic cartoon that he believed was was super important to understanding animation.”
If studying field history feels tedious, that is a signal you are not genuinely passionate about the field and may not be in the right career lane.
“if that sounds tedious, it's probably not the right lane. Like if it's tedious to learn that this isn't a passion like you're not in the right lane.”
Product instinct—the ability to think about product development naturally—is a rare founder trait that's very difficult to teach, making hire-ability for product-first thinking more important than many investors recognize.
“I think product instincts is another one that that that comes partially from understanding the new edge which we already talked about but like it probably took my whole career for me to fully understand how hard it is to hire someone who's not a product first individual and then get them to be good at it. I'm sure there there are examples, but it's got to be 5% or less of the use case.”
Jeff Bezos's personal criterion for angel investing is simple: would this founder do this no matter what, come hell or high water? He identifies determination as present in all great founders.
“I once asked Jeff Bezos, 'How have you had such a successful angel portfolio? You don't have any free time.' And he says, 'Oh, when I meet an entrepreneur, there's only one thing I ask myself. Is this person gonna do this no matter what? Come hell or high water, they're doing this.”
There is an historical parallel to the AI domination question: Microsoft started with the OS, then climbed the stack to office applications like Word and Excel, displacing competitors like Lotus 1-2-3 and WordPerfect; the same pattern could repeat with AI companies moving from foundational models to vertical applications.
“People point to Microsoft, you know, starting with the OS and then, you know, there were there was Lotus one 123 and there was I forget I can't even Oh, there was a word perfect like I can't remember the the specific apps, but you know, they eventually moved up the stack. That could happen.”
Leaving venture capital as a partner is rare because most partners wouldn't leave unless they felt they had completed their work and had nothing left to contribute.
“I made a decision, a very specific decision to say, 'Okay, I'm done.' And and I I don't think I would have done that if I felt there was work left to do. I think I reached a point where I felt there wasn't any work left to do.”
In the US, if you want to send $50 digitally to someone, you must use ACH which has three-day settlement; wiring money costs $25 and requires forms and possible verbal confirmation, while in Argentina with Pix it is immediate.
“In America, if I want to send you 50 bucks digitally, I've got to go through AC, which is three-day settlement, which is part of this regulatory capture in in Argentina now it's immediate because of pics but you you can wire to you same day but it cost me $25 and I have to fill out a page of forms and I might have to do a verbal commit with my bank”
There was no mentor or case study for the mega-burn level of spending seen in Uber; the board of experienced executives from top companies had never been in this situation before, which created a novel and somewhat terrifying decision environment.
“there was no there was no one to call. There was no there's no mentor to go find. Um which was uh horroring a bit to recognize you're in that situation.”
Matt Kohler resolved Benchmark's website problem by volunteering ownership, then later simplified the site radically to a splash page because any complex site would generate complaints from multiple partners.
“when Matt Kohler came in, he had he's like, 'Oh, man. I'll take it on. I love I know exactly what we need.' He created this super complicated website and it had all the founders on it and now they're connected to all the partners and and people started complaining because stuff wasn't right. And one day, Mac came in and uh he said, 'You know what? I'm taking it all down and I'm putting up a splash page.”
Equal partnership eliminates annual compensation reviews and the political overhead of 'recutting the pie'; economics remain constant so no energy is spent on internal negotiation.
“You don't spend any time annually on comp review and recutting the pie. You don't like it's all it's always equal. it's always going to be equal like that that amount of political overhead just goes away.”
As a venture capitalist, maintaining five premium AI accounts simultaneously is necessary to avoid missing emergent applications and staying ahead of the competition.
“I have like five premium AI accounts cuz I just don't want to miss something. and you get trained that way. I think everyone should operate that way.”
The future of AI models is uncertain; models could consolidate into one dominant general-purpose model that does everything, or specialized vertical models could persist, or we could see continued swaps between models as optimization and price become the primary objective functions in a few years.
“Do you think we're going to end up with like one model that just sort of like dominates or do you think we're going to end up with niche models and they're effectively going to be commodities in some way? I think it's highly dependent on how things play out. There are certain examples in the verticals especially in the coding one which is probably the largest vertical right now where people have swapped out models you know cursor even lets the the user pick the model that they're using and as we move towards optimization and price optimization which isn't really the objective function right now but it will be in a few years um you may see more people try and do those swaps.”
One of the most powerful solutions to improving AI models is hiring expert humans for thousands of dollars per hour to fine-tune models and ask hard questions, creating a human-in-the-loop improvement process.
“One of the most powerful um solutions to improving the models is hiring experts. literally hiring experts for thousands of dollars an hour to to sit in and fine-tune and and you know ask very hard questions and then tune them to to be able to solve those.”
Robinhood announced it would enable tokenized private company equity secondary trading, and companies immediately pushed back arguing the practice would be illegal, indicating regulatory uncertainty and legal resistance to disrupting current equity markets.
“Robin Hood announced they were going to do what you just said and and the companies, you know, threw up a a strong argument that that would be illegal, like you don't have a right to do that. So, we'll see how that plays out.”
Gurley discovered a passion for storytelling and long-form narrative through business school when reading business books, personal development books, then biographies, and finally long-form non-fiction journalism by writers like Malcolm Gladwell, Michael Lewis, and John Crockower.
“I started reading and I I started with business books that most people know. Um, I got into personal development books, which I find a lot of successful people have this moment in their life where they roll through, you know, Dale Carnegie and like like Seven Habits and stuff like that. Um, and then and then biographies. Um, but but after that, I I I've kind of fell in love with with long form non-fiction journalism that reads in an exciting way.”
Magnus Carlson, the world chess champion, won a chess trivia contest during a World Chess Tournament break, demonstrating that mastery of field history is characteristic of top performers.
“There's a, you know, world chess tournament and they take a break and run a trivia contest and Magnus Carlson wins the trivia contest and it's all about the history of chess.”
Examples of great founder communicators include Jeff Bezos and Toby Lütke (CEO of Shopify); they are exceptionally gifted at describing what they are trying to build, and this skill is valuable in attracting talent and support.
“Um and you can see it with Bezos. You can see it with like Toby at Shopify. I mean, God, like, listen to any Toby podcast you possibly can. Like, of course, the world's going to follow this guy, Daniel E. Like, they're just so gifted at describing what they're trying to do, you know, and that's that's just just, you know, super super valuable.”
Venture capital was Gurley's dream job—work so intrinsically satisfying that he would do it for free or minimal salary, and he has no regrets about his career choice.
“that was my dream job. I was thrilled to do it. I loved every minute of it. I often said that I would if we lived in a socialist society and I everyone had to work for free, I would still take that job.”
There is 'a little bit of bespoke confidence' in having just a splash page; it suggests the company doesn't need to explain itself because the work speaks for itself.
“There's a little bit of bespoke confidence in in just having a splash page. I would just add that there are plenty of highly successful venture firms that aren't structured that way.”
Gurley plans to apply techniques that made him successful in venture capital—specifically writing, understanding problems, and synthesizing ideas—to larger societal problems with the goal of making meaningful impact.
“I would like to take some of the techniques that I use to be successful as a venture capitalist mostly around the blog and understanding problems and synthesizing and see if I can apply those techniques to bigger broader problems in society and see if I can dent the universe a little bit that way.”
Gurley read foundational investing books early in his career—Peter Lynch's One Up on Wall Street, Burton Malkiel's A Random Walk Down Wall Street, Warren Buffett's letters, Benjamin Graham's works, and Howard Marks' writings—which provided him with a strong bedrock of financial understanding that later enabled him to innovate on top of a solid foundation when transitioning into venture capital.
“I started on Wall Street, you know, and not in venture directly, I got caught up in all the people you would expect, you know, around Wall Street and stocks. And so, you know, that starts with Peter Lynch, One Up on Wall Street, you know, best-selling book, probably the first book I read about investing, a random walk down Wall Street, Burton Male, all the Buffett letters, you know, uh, Ben Graham. Once you read Buffett, you have to read Ben Graham.”
Different AI models have different strengths: ChatGPT has good project structure and memory features; Gemini has access to Google review data and allows deep menu exploration; Claude is preferred by coders; Perplexity is preferred for finance; and Claude is better for deep research on companies and countries.
“I I like the project structure and I'm being sucked into the memory element in that it knows who I am and it knows things about me for for restaurants and stuff I' I've been using Gemini just because it has all the Google review data and you can, you know, you you don't just ask it which restaurants are good. You can say what are three plates people rave about and what have people warn against. Like you can go deep into the menu, which I I do all the time. You know, the coding people swear by uh Claude.”
There are many successful VC firms with different structures than Benchmark's equal partnership; the structure is not the only way to succeed, just one effective approach among multiple viable models.
“I would just add that there are plenty of highly successful venture firms that aren't structured that way. And so I don't it's not I'm not saying it's the only way to do it. It's clearly there clearly many ways to do it”
AlphaGo learned through self-play, training itself by playing millions of games against itself without explicit human instruction, demonstrating learning from pure competition rather than supervised training.
“And that was trained just by playing. Is that true? Yeah. Yeah. Yeah. Exactly.”
Gurley would be comfortable sitting in the back seat of a Tesla with full self-driving enabled and would not feel a need to drive anymore; he finds FSD 'scary good.'
“it's scary good [laughter] I mean I was telling someone the other day I was like I would be comfortable sitting in the back seat at this point with full self-driving like I don't I don't feel a need to drive anymore.”