Y Combinator CEO Garry Tan: Turning Ambitious Misfits into Founders

Apr 29, 2025 Episode Page ↗
Overview

Garry Tan, President of Y Combinator, discusses what makes YC successful, the critical trait of earnestness in founders, and how AI is transforming startups and venture capital. He shares insights on building, focus, and the future of technology and innovation.

At a Glance
25 Insights
2h 20m Duration
25 Topics
7 Concepts

Deep Dive Analysis

The Genesis and Success of Y Combinator

Y Combinator's Program and Application Process

The 10-Minute Interview and Identifying Founder Potential

Y Combinator's Unique Success Rate and Long-Term Outcomes

The Transformative Environment of YC and Founder Focus

San Francisco's Role in Innovation and Ideal Policies

Ethical Considerations in Funding Startups

The Critical Importance of Sales Mindset for Founders

Earnestness as a Key Trait for Successful Founders

The Shifting Landscape of AI Companies and Blitzscaling

Incumbents vs. Startups in the AI Era

Navigating AI Regulation and the Concept of Agency

The Future of Work: Robotics, UBI, and AI-Driven Progress

AI's Role in Scientific Discovery and Manufacturing

Data Access, Patentability, and AI Outputs

Concerns and Optimism Regarding AI's Future

Meta's AI Strategy and the Evolution of Computing Interfaces

Defining AGI and the Golden Age of Building AI

Effective Prompting and Competitive Advantage in AI

Investment Strategy in Big Tech AI Companies

Lessons from MrBeast on YouTube Optimization

Paul Graham's Influence on Clear Communication and Founder Mindset

The Remarkable Reality of Startup Success and Failure

Sam Altman's Impact and the Future of Humanity

Defining Personal and Societal Success in the Age of AI

Shelling Point

A concept where a focal point or common reference allows individuals to coordinate their actions without explicit communication. Paul Graham's essays served as a shelling point, attracting like-minded individuals to Y Combinator who shared his vision for building software.

Founder Mode

A state where the founder and CEO actively exercises agency and plays a direct role in all aspects of the company, rather than delegating entirely. This approach counters the common management advice of hiring the best and giving them maximum autonomy, which can lead to internal politics and disempowerment in larger organizations.

Earnestness

A core trait for successful founders, characterized by sincerity, authenticity, and humility. It means being genuinely focused on solving a real-world problem rather than pursuing external validation like money, fame, or status, and is seen as a driver of durable success.

Blitzscaling

A strategy for rapid growth, often seen in the 2010s with companies like Uber, where access to capital was used to scale exponentially and dominate markets. In the AI era, this concept may be less relevant as companies achieve significant revenue growth with far fewer people due to the leverage of AI models.

Test Time Compute

A method where an AI model spends more time at the query level, essentially 'thinking' longer before providing an answer. This increased processing time, even if it means a delay of minutes, can lead to significantly more correct and reliable outputs, especially for complex tasks.

Synthetic Data

Data generated by an AI model itself, which can then be used to train and improve the same or other models. This process allows models to self-bootstrap and potentially accelerate their learning and development.

Golden Evals

A set of rigorous test cases used to evaluate the performance and accuracy of AI models, particularly in specific domains. These evaluations, often developed by expert prompt engineers, act as a 'moat' for companies, ensuring their AI applications deliver reliable and deterministic outputs.

?
What makes Y Combinator so successful?

Y Combinator's success stems from attracting ambitious founders through Paul Graham's essays, providing a 10-week program that transforms worldviews, offering capital and know-how, and fostering a community focused on earnest building rather than status-seeking.

?
How does Y Combinator filter tens of thousands of applications with such a high success rate?

YC uses a combination of software and human review by 13 general partners who read applications and watch one-minute videos. The final filter is a 10-minute interview focused on crisp communication and understanding if founders genuinely grasp the problem they're solving and why they're working on it.

?
What is the typical success rate for Y Combinator companies?

About 5.5% of YC companies become unicorns (valued at over $1 billion), with some batches reaching 8-10%. Approximately 2.5% become decacorns, and about half of all YC companies eventually raise a Series A funding round.

?
Why is San Francisco considered a unique hub for innovation, and what policy changes could enhance it?

San Francisco acts as a 'shelling point' for ambitious, techno-optimistic individuals, with teams based there doubling their chance of becoming a unicorn. To enhance this, policies should focus on radically increasing housing supply to lower rents and make the city more livable for builders.

?
What is the biggest unexpected change in building companies in the AI world?

The biggest change is that 'blitzscaling' may no longer be necessary; AI companies are achieving rapid revenue growth (e.g., $0 to $6M in 6 months or $0 to $12M in 12 months) with very small teams (under a dozen people) due to intelligence being 'on tap'.

?
How should AI be regulated, or should it be regulated at all?

Regulation should focus on fostering competition and open systems, rather than premature restrictions based on 'Terminator 2' fears. The current AI systems lack agency, and existing laws often cover concerns like bioterror. The focus should be on ensuring AI increases the quality of life for all and promotes human agency.

?
How can one create a competitive advantage in the rapidly evolving AI landscape where models are constantly improving and becoming open source?

The competitive advantage in AI is not necessarily the model itself, but rather the 'golden evals' (rigorous test cases) and superior user experience. Building great software with intuitive UI, strong sales, and customer retention remains crucial, even if fewer people are needed to achieve it.

?
What is the most important lesson Garry Tan learned from Paul Graham?

The most important lesson is to be plain-spoken and hyper-aware of artifice or 'bullshit.' This involves focusing on high signal-to-noise communication, spending time on clear thinking (writing is thinking), and being able to articulate complex ideas concisely, such as in a two-sentence pitch.

1. Cultivate Earnestness and Sincerity

Be incredibly sincere, authentic, and humble in your endeavors, as this trait correlates with durable success and is the opposite of superficial ‘hacks’ or affectations.

2. Focus on Solving Real Problems

Start a company or project with the primary goal of solving a real problem in the world that you deeply understand and believe technology can address, rather than just wanting to be a founder.

3. Prioritize Long-Term Value Creation

Concentrate on building real enterprise value and solving problems over 5-15 years, rather than chasing short-term popularity, status, or superficial metrics like social media followers.

4. Maintain Unwavering Focus

Avoid distractions like chasing social media followers, excessive fundraising, or ‘high-status’ activities, as taking your eye off core building can lead to startup failure.

5. Embrace Low-Status, Early Work

Recognize that significant future value often originates from early-stage, ’low-status’ work deep in the weeds, so be willing to engage in fundamental, unglamorous tasks.

6. Adopt Aggressive Sales Mindset

Overcome the fear of rejection by actively and quickly seeking ’no’ from potential leads, allowing you to efficiently move on to more promising opportunities.

7. Break Down Complex AI Prompts

When using AI models, divide complex tasks into smaller, sequential steps within prompts to reduce hallucinations and achieve more deterministic, reliable outputs.

8. Model Human Workflows for AI

To effectively leverage AI for knowledge work, meticulously model how a human would perform a task, break it into discrete steps, and create evaluations for each step.

9. Build Durable AI Moats

Understand that the underlying AI model is not the moat; instead, focus on developing proprietary ‘golden evals’ (test sets) and a superior, intuitive user experience.

10. Anticipate Continuous AI Model Improvement

Always assume that AI models will rapidly improve, and design your strategy to adapt to these advancements rather than relying on current model limitations as a long-term competitive advantage.

11. Practice Plain Spoken Communication

Strive for direct, clear communication and be vigilant against artifice or ‘bullshit’ in your own and others’ messaging, as it reflects clear thinking.

12. Master the Two-Sentence Pitch

Develop a concise two-sentence pitch that clearly explains ‘what it is’ and ‘why it’s important,’ as it’s crucial for communicating your vision and attracting support in brief encounters.

13. View Writing as Thinking

Dedicate significant time to iterating, whittling down words, and combining complex concepts into concise language, recognizing that this process is a form of deep thinking.

14. Cultivate Multi-Level Understanding

Develop the ability to understand a topic at various levels of detail, from high-level strategy to granular specifics, without getting frustrated by complexity.

15. Surround Yourself with Earnest Builders

Seek out and engage with communities of people who are genuinely and earnestly trying to build things, as this environment can significantly amplify your own efforts.

16. Learn from All Outcomes

Continuously analyze both successes and failures in your selection and decision-making processes to refine your judgment and improve future outcomes.

17. Embrace Geographic Concentration

Increase your chances of success by being physically present in established innovation hubs like San Francisco, where the culture of building is pervasive.

18. Prioritize Ethical Considerations

Be willing to decline profitable opportunities if they conflict with your ethical principles or could have negative societal consequences, even if they are likely to make money.

19. Foster Agency in Children

Design educational experiences that actively increase children’s agency and decision-making skills, such as through interactive games and open-ended play, rather than systems that take agency away.

20. Advocate for Open Systems, Choice

Support policies and technologies that promote open systems and consumer choice, as these are fundamental to liberty and human flourishing.

21. Design for Continuous Innovation

Implement mechanisms (e.g., regulatory frameworks) that prevent complacency and encourage ongoing innovation by automatically eroding advantages over time.

22. Maintain Human Oversight with AI

Even with highly capable AI systems, ensure a human remains in the loop to exercise judgment and intervene when necessary, similar to an autopilot with a pilot.

23. Optimize YouTube Titles for Clicks

Use clickbait titles for YouTube videos to improve click-through rates, but ensure they are relevant enough to the content to maintain watch time.

24. Optimize YouTube Thumbnails Visually

Create distinct and recognizable YouTube thumbnails, often featuring a person looking at the camera and a consistent style, to improve brand recognition and click-through.

25. Shamelessly Ask for YouTube Engagement

Actively ask viewers to like, subscribe, and especially hit the bell icon, as notifications are the most effective way to ensure your content reaches them.

If in 10 minutes you cannot actually understand what's going on, it means the person on the other end doesn't actually understand what's going on. And there isn't anything to understand, which is surprising.

Garry Tan

If you're a startup founder and suddenly, you know, people have heard of you and people try to add you as a scout. Like people kill their startups all the time by that just by taking their eye off the ball.

Garry Tan

The default startup scenario out there is not about signal. It's about the noise. Like, you're playing for these other things, like, how much money can I raise and what, you know, high-status investor.

Garry Tan

The things that actually win. You know, I mean, and going back to Buffett, you know, I went to their, you know, sort of conclave in Omaha. I mean, amazing. And I think those guys are, by definition, extremely earnest.

Garry Tan

The world is full of problems. Like why are people sort of retired in place pulling down, you know, insane by average American standards, absolutely insane salaries to build software that, you know, doesn't change, doesn't get better.

Garry Tan

It's like, these are like extremely influenceable systems. Your idea might be best, but I'm going to disagree because it's your idea, not my idea.

Garry Tan

The second best software in the world for everything is using ChatGPT because you can basically copy and paste, you know, almost any workflow or any data. And it's like the general purpose thing that, you know, you can just drop data into it. And it's the second best because the first best will be a really great UI made by a really good product designer who's a great engineer, who's a prompt engineer, who actually creates software that doesn't require copy-paste.

Garry Tan

This is the golden age of building.

Garry Tan

Sales Mindset Shift for Founders

Garry Tan (referencing Spencer Skates of Amplitude)
  1. Recognize that sales is hard and often unnatural.
  2. Adopt the mindset of 'don't run away from the no'.
  3. Aggressively try to get to a 'no' from potential leads.
  4. Move on quickly from 'no's to spend less time on unproductive leads and find the 'gold nuggets'.

Effective AI Prompting for Complex Tasks

Garry Tan (referencing Jake Heller of Case Text)
  1. Identify a complex task that a human knowledge worker would perform.
  2. Break down the complex task into smaller, more specific, and granular steps.
  3. For each micro-step, create a distinct prompt for the AI model.
  4. If the AI model hallucinates or fails, it indicates that the prompt is asking it to do too many things; break that step down further.
  5. Develop 'golden evals' (test cases) for each step to ensure deterministic and accurate output from the AI.

YouTube Channel Optimization (MrBeast's Advice)

MrBeast (as recounted by Garry Tan)
  1. Create clickbait titles that grab attention.
  2. Design thumbnails that feature a person looking into the camera.
  3. Ensure thumbnails are highly recognizable and consistent in style.
  4. Focus on watch time as the most crucial metric for YouTube's algorithm.
  5. Shamelessly ask viewers to like, subscribe, and hit the bell icon, as bell notifications are the most effective way to reach subscribers.
1%
Y Combinator acceptance rate Out of 70,000-80,000 applications annually.
60%
Unicorn startups from YC alumni (last decade) Percentage of all unicorn startups in the last decade that had YC alumni as founders.
10 weeks
Y Combinator program duration Length of the YC program.
$1 million to $1.5 million
Median funding raised by YC companies post-Demo Day For teams of two or three people, sometimes starting with just an idea.
$1 billion
Annual funding into YC companies Total funding received by YC companies per year.
$500,000
Initial funding amount offered by YC today Increased from $20,000 at the start.
5.5%
Unicorn rate for YC companies Percentage of YC companies that become billion-dollar companies (up from 3.5-4% 10-15 years ago).
2.5%
Decacorn rate for YC companies Percentage of YC companies that become companies valued at $10 billion or more.
25%
YC companies raising Series A in year 5 or later Of those YC companies that raise a Series A.
40%
Unicorns started by multi-time serial founders (last 10 years) Percentage of all unicorns worldwide.
60%
YC alumni among multi-time serial founders who created unicorns Percentage of the 40% mentioned above.
$0 to $6 million
Revenue growth for AI companies (example 1) In 6 months with under a dozen people.
$0 to $12 million/year
Revenue growth for AI companies (example 2) In 12 months with under a dozen people.
$200/month
Cost of 01 Pro (test time compute) A service for enhanced AI query processing.
84%
AIME math test score for a 1.5 billion parameter model Achieved by DeepSeek R1, a model small enough to fit on a phone.
$4 billion
Money sloshing around AI safety organizations Estimated amount, leading to potential premature justification of activities.
80%
Customer support volume handled by AI for wine merchants Ordering volume handled with no human in the loop for some customers of a YC-funded company.
1%
Garry Tan's PDOOM score (probability of doom from AI) Indicates a low but non-zero level of concern about existential risks from AI.