Paul Graham on the Essence of Entrepreneurship: YC Batch 47 Training, Extreme Ambition, and the Portrait of Founders of Trillion-Dollar Giants

This Paul Graham interview, I believe, is much more important than it appears on the surface. Because PG is not doing a "nostalgic entrepreneurship interview." He is actually answering a very big question: After AI has re-priced coding, product development, knowledge acquisition, and even research capabilities, what aspects of entrepreneurship remain unchanged by technology? His answer is very clear: The tools have changed, but the underlying humanity of founders has not. On September 3, 2026, YC released this interview of Paul Graham by Vivian Shen at the original YC office in Mountain View. PG said this was about the 47th founder talk he has given in 21 years facing a YC batch; even though he has spoken dozens of times, he still starts with a blank text file to rewrite before each speech, because the entrepreneurial environment changes, but the vast majority of entrepreneurial principles remain very stable. And the entire interview can really be summarized in one sentence: The hardest asset to replicate for future trillion-dollar companies is not the idea, but the Formidable Founder. ──────────────── 1. Upgrade the title first: The core word in this article should be "Formidable" Your current title direction is basically correct, but if I had to choose, I would shift the focus from "the 47th internal training" to PG's most important talent judgment. Precision type Paul Graham's 21 years of entrepreneurial observation: After 47 YC batches, why do "Formidable Founders" still determine the birth of trillion-dollar companies? Trend type AI changes everything, but entrepreneurs have not changed: Paul Graham talks about Formidable Founders, rapid delivery, and the next trillion-dollar giant. Entrepreneurship type When the server crashes, no one thinks about becoming a billionaire: Paul Graham dissects the true driving force of great founders. My top recommendation In the AI era, what hasn’t changed? Paul Graham explains the underlying code of great founders using 21 years of YC samples. This title has the highest level. Because what this content really talks about is: Invariants of Entrepreneurship That is: The invariants of entrepreneurship. ──────────────── 2. First, correct "the 47th training camp": The accurate statement is PG's approximately 47th YC batch PG himself said: Over the past 21 years, he has probably given similar founder talks for about 47 batches. It is not a fixed course called: "YC's 47th Entrepreneurship Class" And he still: Starts preparing from a blank document each time. This is very interesting. Why? Because true experts do not copy: "I have given this talk 46 times" and just replicate: Talkv46Final_Final2.pptx. He knows: Principles remain stable. But: Context changes. ──────────────── 3. This is actually a very advanced cognitive method It can be called: Stable Principles × Updated Reality For example: Entrepreneurship in 2005 required: Doing what users want. In 2026: It still requires that. In 2005: Fast shipping was important. In 2026: It is still important. In 2005: Founders must be resilient. In 2026: It is still so. But: The main costs for startups in 2005 might have been: Employee salaries, Servers. In 2026, many AI startups have begun to see significant costs emerge: Token / GPU / inference spend. PG even specifically mentioned in this interview: Now some AI startups' token bills have become cost items on par with or even larger than salaries. So the principle hasn’t changed: Control Burn. But: The structure of Burn has changed. This is why he rewrites. ──────────────── 4. This is the thinking that entrepreneurs should learn the most: Do not mix "eternal laws" with "current practices" Many people learn entrepreneurship: By reading articles from 2008. Then they copy: The tools from 2008. Wrong. What should really be extracted is: Principle. Not: Implementation. For example: PG says: Talk to Users. Principle: Get real market feedback as soon as possible. 2005: Make phone calls. 2015: Intercom. 2026: Perhaps an AI agent automatically summarizes thousands of user feedback daily. Tools change. Principles do not. ──────────────── 5. PG's rebuttal to the "golden age of early YC" is actually very interesting There has long been a kind of Silicon Valley nostalgia in the outside world: "The old YC was the real YC." Reddit. Airbnb. Dropbox. Stripe. Now it’s too big. It has changed. PG said: He has been hearing such comments since: 2008 That is to say, about three years after YC was founded, Some people were already saying: "The golden age is over." This actually exposes a very common human cognitive bias: Golden Age Bias. ──────────────── 6. Why does every generation feel that "the past was better"? Because: The failures of the past have disappeared. Only the: Winners remain. Today's startups: Are still in a chaotic state. So you take: 100 unproven companies in 2026 and compare them to: Reddit, Airbnb, Stripe that survived 20 years after filtering from 2005 to 2015. Of course, it feels: The past was more impressive. This is called: Survivorship Bias. ──────────────── 7. Reddit in 2005 did not look like a "historic company" at all Today looking at Reddit: A multi-billion dollar listed platform. Very serious. But in the first batch of YC in 2005: It was essentially just a: Link aggregation website. PG even joked about it in the interview: Back then, Reddit was not some "intercontinental ballistic cargo." What he means is: Today, some YC startups have clearly surpassed early typical internet products in terms of physical complexity and scientific difficulty. ──────────────── 8. But here I must correct the "intercontinental ballistic cargo" in your materials This is an extremely memorable joke name coined by PG. The real company is: Hop Aero (YC S26) The product is called: Rook. The official specifications currently are: Transport: 250 kg Maximum approximately: 750 km About: 15 minutes. It launches from a standard 40-foot container, performs suborbital/hypersonic flight, and lands at unpaved locations. The example YC gave is even Okinawa → Taiwan. So: It is currently not literally "intercontinental" missile logistics. 750 kilometers is obviously not Intercontinental. "Intercontinental Ballistic Cargo" is a humorous name PG gave to this concept. The founder of Hop Aero has also publicly stated that this term was conceived by PG during their office hours. ──────────────── 9. What’s really interesting is not "delivering packages with missiles" But a very advanced entrepreneurial question: When is the value of time high enough to change the mode of transportation? Ordinary Amazon packages: Arrive in two days. No problem. But: Battlefield medical supplies. Satellite parts. Critical industrial components. Offshore platform parts. Aircraft AOG parts. Every hour of delay could lead to: Losses of: Hundreds of thousands or even millions of dollars. One of the origins of the Hop Aero founder's project is: They once had to deliver satellite propulsion hardware to the mission integrator within 24 hours, FedEx and UPS could not guarantee the deadline, In the end, the founder made a: 23-hour round trip for hand delivery. ──────────────── 10. This presents a very typical entrepreneurial opportunity: Extreme Pain × Extreme Willingness to Pay Many entrepreneurs ask: "How many people are in my market?" Not enough. They should ask: How painful is the problem? Assuming there are only: 10,000 customers. But each customer loses: $10M a year due to this problem. That’s a huge market. ──────────────── 11. So Hop Aero should not be understood as "SpaceX delivering packages" The real thesis is: Does ultra-high time-value cargo need a new transportation category between planes and missiles? If the answer is Yes, It is creating: A New Transportation Layer. This is what PG calls: Frighteningly Ambitious. ──────────────── 12. And "Frighteningly Ambitious" is not a term he just proposed in 2026 In 2012, PG wrote the famous article: Frighteningly Ambitious Startup Ideas The core meaning is not: Entrepreneurs should shout all day: "I want to change the world." But rather: Real huge opportunities are often so big that: Smart people instinctively shrink back. Because your first feeling will be: "Why should I be able to do this?" One of the biggest categories PG mentioned back then was: Rebuilding Google. ──────────────── 13. But PG's judgment on the "new Google" was not primarily about predicting search But rather: Do not copy the current giants head-on. This is a very important lesson for entrepreneurs to take away from the entire interview. Assuming in 2012 you said: I want to defeat Google. And then do: A search box 10 blue links Better PageRank. You basically have no chance. Because you are in: Google's axis of competition Compete with Google. ──────────────── 14. To truly disrupt a giant, you need to: Change the Axis. PG's meaning back then was roughly: Wait for a certain technological change to occur, Making the original Google model no longer the best way for users to obtain information. More than a decade later: LLM has arrived. Users are starting to: Ask a question ↓ Get an answer directly. Instead of: Ask a question ↓ 10 links ↓ Open 5 web pages ↓ Synthesize it themselves. In this interview, PG clearly views OpenAI as a realization of that "new Google" idea in some sense. ──────────────── 15. However, it must be corrected: "OpenAI is a YC incubated company" is inaccurate. OpenAI is not a typical YC startup that participates in a YC Batch. It does have a very deep connection with YC's system: Sam Altman was the president of YC at the time. In 2015, YC announced the establishment of YC Research. Subsequently, YC officially referred to OpenAI as: "the first group affiliated with YC Research." However, OpenAI ultimately became an independent non-profit organization. The internal plan at that time considered making the AI lab more directly associated with YC, but it was later decided that OpenAI should operate independently of YC; OpenAI has publicly shared relevant emails and history from that time. So the most accurate statement is: OpenAI has a deep founding connection with Sam Altman and YC Research, but it is not a traditionally defined "YC incubated graduate company." ──────────────── 16. PG's true insight about Google can be elevated to a startup iron law: You rarely kill an incumbent by becoming a slightly better incumbent. You seldom rely on: Creating a good 10% product to kill a giant. Real disruption comes from: Paradigm Shift. Kodak: Was not killed by better film. But by: Digital Camera. Blockbuster: Was not killed by another video rental store. But by: Streaming. Taxi: Was not killed by a better taxi company. But by: Smartphone + Marketplace + GPS. Search: The real challenge it faces is not: Another Search Engine. But: Generative Answer Engine. ──────────────── 17. This explains why "extreme ambition" and "starting from something small" are not contradictory This is a point that many entrepreneurs easily confuse. Vision: Can be extremely large. Initial product: Should be extremely small. Amazon: Everything Store. First step: Books. Facebook: Connect the world. First step: Harvard. Airbnb: Global accommodation network. First step: A few air mattresses. ──────────────── 18. So the truly advanced startup strategy is: Ambitious Destination + Narrow Wedge Not: Doing everything from day one. If Hop Aero really wants to establish global ultra-fast logistics: The first market might be: Defense, Critical goods, High time-value logistics. That’s correct. ──────────────── 19. Looking again at the cancer part: PG really mentioned Omanta this time Currently, there is a company worth paying attention to in YC Summer 2026: Omanta It is not a vague: "Cancer AI assistant." But aims to: Establish a research lab for one for: Complex cancers, Rare diseases, Undiagnosed conditions. In other words: Provide a patient with research capabilities close to a complete research project level. ──────────────── 20. Omanta's core thesis is very bold The current medical system is: Doctors care for a large number of patients at the same time. Even for the most complex cases: Clinical doctors cannot possibly read full-time: Every paper, Every experimental treatment, Every biomarker, Every combination scheme. Omanta wants to consolidate: Medical records, Genomics, Multi-omics data, Papers, Treatment candidates To establish a continuous research system for: One patient. This is what is called: Research Lab for One. ──────────────── 21. This actually represents a very important trend of AI in healthcare The economics of the past medical system was: Scarce Expert Attention. Doctors' time is limited. Researchers are limited. Complex analysis is expensive. After the emergence of AI: Reading, Data integration, Hypothesis generation, Literature search The marginal cost has decreased. Thus, in the past: Only billionaires or a very few special patients could organize: Personalized Research Teams Theoretically, this may gradually spread. ──────────────── 22. The case of Sid Sijbrandij is indeed one of the important starting points for Omanta GitLab co-founder Sid Sijbrandij was diagnosed with osteosarcoma in 2022. After the cancer recurred, As standard treatment options gradually ran out, He began to apply the methodologies from his startup to his medical decision-making: Establishing a single source of truth, Aggregating large amounts of data, Organizing scientists from different institutions, High-frequency testing, Rapid hypothesis formation, Using multi-omics and AI-assisted research. He later referred to this approach as: Founder Mode on Cancer. ──────────────── 23. But it must be very cautious: it cannot be written as "entrepreneurial methods cured cancer" This is a false causation. Sid currently publicly states: no evidence of disease / complete remission And indeed received very complex, personalized medical and experimental interventions. But this is still: N=1. It cannot simply prove that: A certain set of Founder Mode methods Can universally cure cancer. It involves: Medical teams, Experimental therapies, Disease biology, Individual responses, Timing And many other variables. So the most professional expression is: Sid's experience demonstrates the potential value of high-density personalized research in extremely complex cases, rather than proving a universal "cancer entrepreneurial therapy." ──────────────── 24. The two founders of Omanta are indeed directly related to Sid's case Alfredo Gonzalez previously deeply participated in Sid's personalized therapeutic research and handled a large amount of genomic and multi-omics data. Ranad Humeidi has established and managed research projects for complex individual patients. YC's company page clearly lists Sid's case as one of the important sources for their formation of Omanta's methodology. So this point is valid. ──────────────── 25. PG's description of cancer as "Death of a Thousand Cuts" is also very thought-provoking The real meaning is not: "One cut solves cancer." But possibly: Multiple paths attacking simultaneously. Diagnostics. Vaccines. Immunotherapy. Targeted Therapy. Radioligands. Monitoring. Continuously obtaining new data, Continuously adjusting. This is an: Adaptive System vs Adaptive Disease. ──────────────── 26. This thinking is actually very similar to entrepreneurship Cancer cells: Adapt. Compete. Mutate. The entrepreneurial market: Also constantly changes. So static plans often fail. What really matters is: Feedback Loop Speed. Observe. ↓ Hypothesize. ↓ Act. ↓ Measure. ↓ Update. This is the commonality between entrepreneurship and complex scientific systems. ──────────────── 27. This also brings out a point in this interview that I think is more important than "trillion-dollar companies," but your raw materials did not emphasize enough: Shipping Pace remains one of the most important success signals for YC. PG said: Even today, with: Claude, Codex, AI coding agents, Massive automation tools, Many YC startups: Still do not ship fast enough. And he believes: The speed at which founders continuously deliver new things, Is still one of the most important signals to predict startup success or failure. This is very important. ──────────────── 28. Why is it that as AI becomes stronger, Shipping Speed becomes even more important? Because tools have been democratized. In the past: An engineer would take two weeks. Now: AI helps you finish it in a day. But: Your competitors also have AI. So: Tool Advantage ↓ In the end, what really makes the difference is: Decision Speed. User Feedback Speed. Experiment Speed. Learning Speed. ──────────────── 29. So AI does not necessarily reduce entrepreneurial competition It is very likely that: AI lowers the cost of product creation but raises the speed threshold for entrepreneurs. In the past: Releasing a feature once a month. Was normal. In the future: Releasing a feature once a month. May already be very slow. This is like: After the invention of the automobile, Not everyone slows down. Rather: The speed standard of the entire world is rising. ──────────────── Thirty, this may be a very important rule in the era of AI entrepreneurship. When production gets cheaper, iteration becomes the bottleneck. After production becomes cheap: The truly limited resources become: Judgment. Taste. Customer Understanding. Prioritization. In other words: The best founders in the future may not necessarily be: The fastest coders. But rather: The ones who complete the Learn → Decide → Ship → Learn cycle the fastest. ──────────────── Thirty-one, now let's talk about one of PG's most famous terms: Formidable Founder This term does not mean: "Very confident." Nor does it mean: "Loud voice." PG defined a formidable person very early in his article "How to Convince Investors": A formidable person is someone who makes you feel "no matter what obstacles are ahead, he will probably get what he wants in the end." This definition is extremely good. ──────────────── Thirty-two, why does "smart" not equal Formidable? A person with an IQ of: 150. May be very smart. But: Afraid of rejection. Does not sell products. Does not hire people. Crumbles at bad news. Cannot negotiate. Thus: Accomplishes nothing. Another person with an IQ of: 130. But: Continues after being rejected by customers 20 times. Changes the product if it doesn't work. Continues after failing to raise funds. Keeps looking for key employees who won't come. In the end: Gets What He Wants. The second person: Is more Formidable. ──────────────── Thirty-three, so Formidable is not a personality trait, but rather "the ability to achieve real-world results." I would write it as: Formidable = Agency × Persistence × Judgment × Adaptation Agency Does not wait for others to arrange. Persistence Does not stop easily. Judgment The direction is not always wrong. Adaptation Can change when realizing mistakes. All four are indispensable. ──────────────── Thirty-four, why is mere "tenacity" still not enough? Assume: A founder: Is extremely tenacious. But always has the wrong direction. Can persist for: 20 years. Still fails. This is called: Stubbornness. A truly formidable person: Has a firm goal. Can change the path. This is: Strong Will + Flexible Tactics. ──────────────── Thirty-five, why is Sam Altman one of the most frequently cited formidable examples by PG? PG mentioned again in this interview: When he first met the young Sam Altman, He felt he was: Extremely formidable. This even happened before YC decided to admit him. What does this indicate? Investors are not really screening for: Resume. But rather: Trajectory. ──────────────── Thirty-six, one of the core signals of Formidable: people actively change their environment. Ordinary people: The environment does not allow. Formidable: Then how can I change the environment? Ordinary people: No one replies to emails. Formidable: Who else can I reach him through? Ordinary people: Customers are unwilling to buy. Formidable: Why? Price? Product? Market? Which one to change? This is called: High Agency. ──────────────── Thirty-seven, this is also why the question "Will founders be replaced by AI?" is actually a wrong question. AI can: Write code. Write emails. Conduct research. Generate ads. Build models. But the real work of a founder is not the sum of these tasks. The work of a founder is: Deciding which things are worth existing, and then making reality conform to that judgment. This is a: Coordination of Reality. AI can assist. But today there is still no evidence to suggest that: This: Ambition + Agency + Taste + Responsibility Can be easily replaced by a model. ──────────────── Thirty-eight, PG's judgment on ambition is also very harsh. His view is basically: Very deep ambition is hard to "teach" later. People can: Be allowed by the environment to show ambition. But the extreme desire to win, the extreme desire to do great things, Exists to a large extent already. This is also one of the reasons why YC prioritizes screening people over "cultivating people." ──────────────── Thirty-nine, this is completely different from the school system. Schools assume: Talent can be standardized and cultivated. YC is more like: First find the outliers, then give them an environment. This is actually very similar to: Professional sports, Music academies, Top laboratories. You cannot train: Anyone To become: Magnus Carlsen. But: If Magnus appears, You can: Train him faster. ──────────────── Forty, so the real core asset of a top Accelerator is not the curriculum. All startup advice from YC: Has long been publicly available for free. YouTube. Essays. Startup School. Anyone can watch. Why is YC still valuable? Because: People + Selection + Network + Intensity. This is the key. ──────────────── Forty-one, you mentioned "Pods of about 70 companies," this needs to be updated. YC is indeed expanding the batch and adopting: Sharding / smaller groups To avoid hundreds of companies being mixed together. The official history also clearly states: The first batch had only 8 companies; Later, when it broke through 100 and 200 companies, smaller groups were used to maintain the depth of partners' understanding of the companies. But currently, the official description of the batch structure by YC is: 3 large Groups Each group is further divided into: Sections of about 6-10 companies. It is not a fixed system of "about 70 companies per Pod." So the original text suggests a change. ──────────────── Forty-two, and this organizational design is very worthy of management research. YC faces: The Dunbar / Scale Problem. 8 companies: Everyone knows each other. 200 companies: Impossible. If you forcibly maintain a single group: Network density rapidly decreases. So what YC is doing is essentially: Shard the Network. ──────────────── Forty-three, this is exactly like distributed systems. One database: Small data is good. When it gets large: Shard. One organization: 20 people: The CEO knows everyone. 2000 people: Must split units. One community: 20 companies: Everyone knows each other. 200 companies: Need subgroups. So organizational design itself is: Information Architecture. ──────────────── Forty-four, the reason YC can expand without completely losing the small group experience is because: Global Network + Local Intimacy. Large network: Has thousands of alumni. Small sections: Have 6-10 companies. This is a very beautiful: Scale + Density Combination. ──────────────── Forty-five, this is the same logic we just studied in The Residency. The Residency: 10-25 people living together. YC: Small sections. Why? Because the community that truly changes behavior is not: 100,000 people on Slack. But rather: High-frequency small groups. The large network is responsible for: Access. The small network is responsible for: Trust. This is a very important distinction in community design. ──────────────── Forty-six, why did YC unexpectedly discover Batch in the first place? This part of history is correct in your direction and very interesting. In 2005, PG, Jessica Livingston, Robert Morris, and Trevor Blackwell started YC. The earliest project was called: Summer Founders Program. PG's public announcement that year clearly stated: It was: "An experimental replacement for the conventional summer job." In other words: College students originally went to: Microsoft, Google for internships. YC said: Why not start a business directly during the summer? ──────────────── Forty-seven, the first batch had only: 8 companies. Among them: Reddit. Loopt. And early projects of Justin Kan and others who later founded Twitch, as well as Sam Altman's Loopt at that time. YC initially provided: About $6,000 for each founder Because it was based on the MIT graduate summer stipend at the time. ──────────────── Forty-eight, and Batch was not designed as a genius system from the beginning. This is particularly interesting. Jessica Livingston later clearly stated: The four of them actually: Had no real angel investing experience. So they thought: "Let’s invest in a bunch at once, so we can learn faster." The results show: Batch is better for entrepreneurs. Because entrepreneurship is originally extremely lonely. Now: A dozen people experience it together. At the same time: YC itself is also more efficient. Once: Talk about financing. Everyone listens. Once: Talk about recruitment. Everyone learns. ──────────────── Forty-nine, this is actually a typical entrepreneurial innovation: Accidental Product-Market Fit. What they really want to solve is: "How to learn angel investing?" The result is an unexpected discovery: Batch itself is a product. This is similar to many great companies. Slack: Originally an internal tool for a gaming team. Twitter: Came from an internal experiment at Odeo. Instagram: Reconstructed from Burbn. Entrepreneurship is often like this: You think you are creating A. Users actually need B. ──────────────── Fifty, so the ability of an excellent founder is not to always be right from the beginning but to: Recognize Unexpected Truth. Is Batch better than individual investment? Then change it. Users only use a small feature in the product? Then cut everything else. Reality tells you: Your original thesis is wrong? Update it. This is called: Reality Responsiveness. ──────────────── Fifty-one, now let's talk about your very exciting material "the model train falling off the table" This metaphor is real, and I think this is the part where PG understands the psychology of entrepreneurs the best. He said: The outside world thinks the founder's motivation is: "I want to become a billionaire." But on some day in reality: The server crashes. Customers are about to churn. Key employees resign. Funding is running out of money. At this time, the founder is not thinking: "Fixing the server brings me closer to a billion dollars." What is in his mind is: "It's over! My train is about to fall off the table, quickly catch it." YC officially paraphrased this part of his words. ──────────────── Fifty-two, this actually reveals a classic economic problem in the motivation of entrepreneurs Long-term incentive: Equity. Short-term motivation: Loss Aversion. Behavioral economics has proven: The pain of loss for a person, Is usually stronger than the pleasure of equivalent gain. Entrepreneurs are the same. Becoming a billionaire in the future: Very abstract. But: The company might die today: Extremely specific. ──────────────── Fifty-three, so why can entrepreneurship allow people to work continuously for ten years? It's not that there is: Positive Motivation every day. More often it is: Local Emergency Resolution. Today: Save the server. Tomorrow: Save the customers. The day after tomorrow: Save the funding. The day after that: Save the recruitment. Continuously saving: For 10 years. Suddenly one day: The company: $10B valuation. This is PG's model train theory. ──────────────── Fifty-four, this can further derive a very important entrepreneurial principle A grand vision is responsible for choosing the direction, while local crises provide daily energy. Vision: Decides why you are on this path. Fear: Keeps you from stopping today. Only Vision: Is easy to talk empty. Only Fear: Is easy to collapse. The strongest founders have: Pull + Push. Vision pulls in front. Failure pushes from behind. ──────────────── Fifty-five, but do not romanticize "fear-driven" This is very similar to Doug Leone's fear-based motivation. Fear is: A strong fuel. But only having fear in the long term: Will lead to: Anxiety, Burnout, Health damage, Poor decision-making. A truly mature founder ultimately needs to upgrade from: Fear of Dying To: Love of Building. Otherwise: After the company succeeds, People still cannot stop running away. ──────────────── Fifty-six, PG also has a very valuable point: entrepreneurship is one of the worst ways to "look cool" This point was particularly highlighted in recent reports about this interview. If a person starts a business just to: Have a good resume, Be respected by others, Look ambitious, Then the ROI of entrepreneurship is extremely poor. Because: Five years later, no one may still know who you are. After failure: There isn't even a clear "certificate." So: Status-Seeking Founder Is usually not stable enough. ──────────────── Fifty-seven, why? Because the rewards of entrepreneurship are extremely: Delayed. Working at a big company: First day: Title. Salary. Logo. LinkedIn. Startup: First day: No one knows you. No money. The product is bad. Customers refuse. So if a person's motivation is mainly: External recognition, The startup's early stage will give him: Negative Status Feedback. It's easy to leave. ──────────────── Fifty-eight, the reward function of a truly formidable founder is often more internalized He does: Not because everyone says: This is a good opportunity. But because: He just wants to see this thing exist. This creates: Intrinsic Obsession. These people are the most dangerous. Because: Even if others do not give rewards, He continues anyway. ──────────────── Fifty-nine, this is also why PG would answer "Where is the next trillion-dollar company?" as: Formidable Founders. And not: AI. Robotics. Biotech. Energy. Space. Because: Ideas are mutable. Founders are: Search algorithms. A strong founder: If Idea A doesn't work, Switch to B. If B doesn't work, Change to C. In the end: Find a huge market. ──────────────── Sixty, this is actually a very deep VC perspective Traditional investment analysis: Invest in Idea. Top Seed VCs: Invest in Search Process. The founder itself is: A Search Algorithm. Excellent founders: Search a larger space. Iterate faster. Correct errors more strongly. So ultimately find: High-quality Ideas with a higher probability. ──────────────── Sixty-one, this is why the Seed stage "the market is still not big" is not necessarily fatal PG's early articles emphasized: The market does not have to be huge right now. You can: Enter from a small market, And then through several hops Enter a larger market. What is really needed is: A credible path to enter: A Big Capturable Market. Airbnb: The initial market was very strange. Uber: Initially black cars. Facebook: College students. None of these were: A trillion-dollar TAM on the first day. ──────────────── Sixty-two, so Founder > Idea should not be misread as "ideas are completely unimportant" This is also a point that must be corrected. A garbage market: Even a strong founder may waste years. The real formula should be: Formidable Founder × Expanding Market × Fast Learning And not: Founder alone. PG's own "How to Convince Investors" also clearly lists: formidable founders, promising market, evidence of success as the three. Only: Founder is the first item. ──────────────── Sixty-three, why are future trillion-dollar companies likely to come from projects that look "very strange" today? Because trillion-level outcomes usually require: New Market Creation. If a market today is already: $500B, Mature giants occupy 90%, You grab 5% from it: The company is good. But becoming: A $1T company is very difficult. Real trillion-dollar companies often: Create: New behaviors, New infrastructure, New platforms. ──────────────── Sixty-four, Microsoft created the PC software platform. Google: The internet information gateway. Amazon: The internet business infrastructure. Apple: The mobile computing platform. Nvidia: The accelerated computing platform. The next one in the future: Is also likely to look like: A market that is not yet fully existing. This is why: PG likes: Frighteningly ambitious. ──────────────── Sixty-five, but huge ideas also bring a danger: they can easily turn into "sci-fi entrepreneurship" A statement: "I want to completely cure cancer." Has no value. A statement: "I want to establish intercontinental rocket logistics." Also has no value. Real entrepreneurship requires: Giant Vision + Next Concrete Milestone. For example, Hop Aero: Not: "Global rocket logistics." But: 250kg. 750km. 15 minutes. Standard container. Test engine. $1.25M US Air Force contract. This turns from: Science Fiction Into: Engineering Program. ──────────────── Sixty-six, this is also a way to judge "real ambition" versus "boastful ambition" Boastful type: Talks about 2050. Real type: Talks about: Next week. Great founders can simultaneously tell you: What the world will look like in 2035, And also: Tomorrow morning at 9 o'clock, which specific blocker needs to be resolved. Both layers have. ──────────────── Sixty-seven, this is why Shipping Pace is such a strong founder signal. Vision is easy to say. Execution is hard to fake. Every week: Is there something new? Is there user feedback? Is revenue moving? Is the product improving? This is the truth. The YC environment is effective largely because: It compresses long-term fantasies into short-term deliveries. ──────────────── Sixty-eight, from an investor's perspective, I would even break down Founder Quality into two axes: Vision Horizon How far can you see? Feedback Loop How fast can you run? The most dangerous person: Sees far, But moves slowly. Turns into: A philosopher. The other: Moves fast, But doesn’t see far. Turns into: A feature factory. The scariest founder is: Long Vision + Short Loop. ──────────────── Sixty-nine, many top founders like Sam Altman, Brian Chesky, Patrick Collison, etc., share similar characteristics. They can discuss: The next ten years. But they also: Dive into specific products, Look at interfaces, Look at users, Look at data. This is also one of the cores of what PG later wrote: Founder Mode. Founders do not automatically just do: "Strategy and delegation" because the company is big. Some key information must be: Directly accessed. ──────────────── Seventy, why did Sid's "Founder Mode on Cancer" resonate? It is precisely because of this logic. When the system's answer is: There are no more standard solutions. He did not just stop at: "Experts will handle it." But instead: Dived into the details. Established: Data systems, Research teams, Feedback mechanisms. Of course, the medical field must be handled by professional doctors and research teams; the proactive nature of entrepreneurs cannot replace medical norms; but what he embodies is: Ownership of the Problem. This is the core psychology of Founder Mode. ──────────────── Seventy-one, PG's entrepreneurial philosophy has always revolved around one word: Agency. Make something people want. Talk to users. Launch fast. Do things that don't scale. Be formidable. Founder Mode. These seemingly different articles, At their core, are all saying: Don't passively wait for the system to solve problems for you. ──────────────── Seventy-two, why has YC been able to last for 21 years? I believe the real answer is not: It invests smartly. But rather, it has established a: Founder Agency Amplification Machine. Screening: High agency people. ↓ Putting them into: High agency peers. ↓ Giving: Funding. ↓ Eliminating: Friction like company registration, financing, etc. ↓ Forcing: Quick launches. ↓ Providing: Market feedback. ↓ Connecting: Alumni networks. It does not create founders. But rather: Amplifies the existing founder traits. ──────────────── Seventy-three, this is very different from universities. Universities: Usually reward: Correct answers. Submitting assignments on time. Meeting requirements. Entrepreneurship: Rewards: Finding problems that no one defines for you. This is also why PG has always liked: Hackers in the early days. Because the underlying spirit of hackers is: "Why must the rules be this way?" ──────────────── Seventy-four, but YC today is far from being a "young hacker summer camp." Winter 2026 alone has about: 199 companies. CB Insights analysis found that about: 1/8 Are working on physical products, With noticeable increases in robots, drones, aerospace hardware, etc.; Industrial & defense companies have also increased significantly. So PG says: Today's projects are "more serious" than in the past. At least from the perspective of technical complexity, there is indeed a basis. ──────────────── Seventy-five, but do not mistakenly deduce that: "Today's entrepreneurs are better than those in 2005." You can't compare this way. In 2005: The internet infrastructure layer. Today: The technology stack is mature. A 22-year-old can call upon: AWS, Stripe, OpenAI, GitHub, Cloudflare, CAD, Simulation software. Therefore today: Two or three people can do what previously required 50 people. So projects appear more complex on the surface, and there is also the effect of: Tool Leverage. ──────────────── Seventy-six, AI is further pushing this leverage to extremes. Previously startups: 10 engineers. Today: Possibly: 3 engineers + agents. Thus: Company Formation Cost Continues to decline. This will produce more companies. However: Good ideas and good founders will not increase proportionally. This is also why the biggest bottleneck in the future is gradually shifting from: Capital And: Coding To: Judgment. ──────────────── Seventy-seven, this connects with all the cases we studied earlier. Brian Singerman: Find Spike. Doug Leone: Find Outlier. The Residency: Increase Talent Density. YC: Find Formidable Founders. You will find that: The stronger AI becomes, Top VCs are increasingly talking about: People. This is not a coincidence. ──────────────── Seventy-eight, because when "making things" becomes cheaper, the biggest difference between people is no longer: Who can write code. But rather: Who knows: What should be done. Who dares: To persist. Who can: Judge. Who can: Hire. Who can: Endure for ten years. These things have not been commoditized with the drop in model API prices. ──────────────── Seventy-nine, so I believe AI will create a very interesting entrepreneurial differentiation. Ordinary abilities: AI Commodity. Extreme abilities: More valuable. Average engineering ability: Value declines. Extreme product taste: Value rises. Ordinary research: Declines. Extreme scientific intuition: Rises. Ordinary content: Declines. Extreme personality and brand: Rises. This completely echoes Singerman's Spike Theory. ──────────────── Eighty, from the perspective of billionaires/capital allocators, I would take away seven key points from this PG interview. First: Don't be superstitious about changes in the era. New technologies will not cancel the basic laws of business. Second: Shipping is the most honest indicator. Vision cannot replace delivery. Third: Truly great markets often do not have a ready-made TAM at the beginning. Companies may create their own TAM. Fourth: Don't directly confront giants on axes defined by them. Look for paradigm shifts. Fifth: Founder is a search algorithm. Ideas will change. People determine how fast they change. Sixth: Fear can drive a company, but mission can sustain it long-term. Seventh: Formidable is more valuable than simply being smart. Those who can push reality towards goals are the ones who ultimately create enterprises. ──────────────── Eighty-one, for entrepreneurs, I would condense it into five practical questions. Ask yourself every day: 1. What did we actually ship this week? Not how many meetings were held. 2. Do users really want it more? Not what the team thinks looks better. 3. What is my biggest bottleneck right now? Solve one at a time. 4. If this direction is wrong, how long will it take me to know? The shorter the feedback cycle, the better. 5. If everything succeeds, how big can this thing really get? Don't just optimize for a small business. ──────────────── Eighty-two, there is also a very important insight to add to your manuscript: AGI is not a clear finish line. PG himself studied AI in the 1980s. In this interview, he said he originally thought the development of intelligence would be like: From: Fly-level, Gradually becoming more complete, And finally crossing: AGI. But the real world is not like this at all. AI has instead shown a kind of: Jagged / Smeared Intelligence. In some areas: Far surpassing humans. In some very ordinary problems: Still making foolish mistakes. He described: AGI is not a fine line, But more like: An area that has been smeared. ──────────────── Eighty-three, this judgment is extremely important for entrepreneurship. Many entrepreneurs are waiting: "Waiting for AGI to come before starting a business." There may not be a clear date for that at all. The reality is more likely: Capabilities break through piece by piece. Code: Breakthrough first. Math: Breakthrough. Video: Breakthrough. Robotics: A bit slower. Social judgment: Another curve. So entrepreneurial opportunities come from: Capability Frontier moving unevenly. ──────────────── Eighty-four, where capabilities have just crossed the threshold of commercial availability, companies may emerge. Voice AI: Suddenly available one year. Coding Agent: Suddenly available one year. Physical AI: Gradually available. Personalized cancer research: Data processing costs decreasing. This is: Why Now. Real entrepreneurs do not wait for AGI. But rather: Focus on: Which capability has just become usable? Eighty-five, this is also the best explanation of "technological change, unchanged entrepreneurial logic". Technology determines: What's newly possible? Entrepreneurial logic determines: Which possibility people actually want? The two must be combined. Only technology: Research projects. Only demand: Traditional businesses. Technology suddenly opens a new space + Founder finds high-value demand: Startup. ──────────────── Eighty-six, so what PG is truly remarkable at is not predicting OpenAI. The long-term principle he insists on is: Great startups live at the edge of what just became possible. Viaweb: The web was just emerging. Reddit: Social web. Airbnb: Internet trust + payments. Stripe: Internet commerce. OpenAI era: AI. Hop Aero: Low-cost aerospace + autonomy + defense demand. Omanta: AI + genomics + personalized research. The underlying principle is: Capability Shift → New Company. ──────────────── Eighty-seven, and why does PG refuse to give an industry answer to "Where will the next trillion-dollar company come from?"? Because industry predictions easily turn into: Consensus Trade. All VCs know: AI is big. Robotics is big. Healthcare is big. This has no edge. The real investment opportunity is: Someone: 22 years old, 25 years old, 30 years old suddenly bringing something that looks very strange. You must judge: Is this person formidable? This is the real difficulty of Seed Investing. ──────────────── Eighty-eight, looking at history, trillion-dollar companies in their early days almost never looked like "future trillion-dollar companies". Apple: Garage computers. Amazon: Online bookstore. Google: Graduate search project. Meta: College social website. Nvidia: PC graphics chips. The real future value: Is usually not fully revealed on: The first pitch deck. So: Founder Quality is an important prior in early uncertainty. ──────────────── Eighty-nine, this also explains why PG ultimately says: great Founders 20 years from now may be very similar to those from 20 years ago. The technological tools will be completely different. But they may still be: Stubborn. Fast. Extremely self-driven. Unwilling to wait for permission. Obsessed with problems. Able to attract others. Scared by failure, fixing their "model trains" every day. Then: Continuously creating things. This is: Human Constant. ──────────────── Ninety, but I would add another layer to PG's conclusion. The real change in the next 20 years may not be: Founder Psychology. But rather: Founder Leverage. 2005: 2 excellent Founders could run a small internet company. 2026: 2 excellent Founders + AI Agents can do what a 20-person company did in the past. After 2030: The organizational leverage of a formidable Founder may be even more exaggerated. So: The economic value of excellent Founders may not decline, but further increase in the AI era. ──────────────── Ninety-one, this is a very important conclusion for the capital market. If: AI increases the productivity of the average person by: 3 times. And increases the productivity of the best entrepreneurs by: 10 times. Then: AI will not narrow the talent gap. Instead: It will amplify the talent gap. This is called: Complementarity Effect. Technology does not replace the strongest. But gives the strongest: Greater leverage. ──────────────── Ninety-two, so the core of future VCs may increasingly not be about "picking companies" but rather: Finding people with extreme leverage capabilities. This relates to: YC's Formidable. Singerman's Spike. Leone's Outlier. In fact, it is an increasingly unified talent theory in Silicon Valley. ──────────────── Finally, I would compress this entire Paul Graham interview into one sentence: The surface technology of entrepreneurship will be completely rewritten every decade, but the human operating system that truly builds great companies has hardly any upgraded versions: seeing problems that others do not see, making things faster than others, quickly changing paths when reality tells you that you are wrong, instinctively reaching out to catch the company when it is about to fall off the edge of the table, and continuously doing this for ten years. And PG's answer to the next trillion-dollar company is essentially not: AI. Cancer. Rockets. Robotics. But rather: Formidable People. Because an Idea is a point in time. A Founder is a continuously operating search system. After technological changes: An Idea can become obsolete. A product can pivot. A market can be reselected. But if that person possesses: Extreme agency, judgment, speed, ambition, and resilience, They may still continue to find the next path. This also explains the most worthy area of study in YC after 21 years: It has never really bet on entrepreneurial ideas, but rather on "what kind of person will ultimately align reality with their goals." This is the highest level of insight from Paul Graham's interview.