Oct 17, 2025 · 49m · mixergy
#2281 Garry Tan: Y Combinator Startups Growing 5X Faster – Here’s What Changed
0:00 / 0:00
gold bands on the timeline = statements, start to end. Hover to read, click to jump. CC turns on captions
Y Combinator President and CEO Garry Tan joins Andrew Warner to discuss how artificial intelligence is driving unprecedented startup growth, the defensibility of vertical SaaS, and the institutional restructuring of Y Combinator.
How this conversation actually went
Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. Andrew holds 24.7% of the talking time here. How this is scored →
Andrew as informed peer 4.8 Guest teaching 4.8 Guest disagreement 1.6 Andrew pushing back 2.6
speaking balance: gold is Andrew, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 24:40 Tan labels mid-career big tech employees as AI deniers
Tan rejects conventional career stability narratives, characterizing ZIRP-era hires clinging to big tech roles as a lost generation in denial about the AI shift.
Hardest push from Andrew ▶ 41:10 Warner presses Tan on why the Continuity Fund was dismantledWarner directly challenges Tan to explain why YC couldn't maintain long-term partner relationships while simultaneously running a growth fund, forcing Tan to defend his bureaucratic simplification.
Biggest teaching moment ▶ 4:32 Tan explains log-linear scaling laws vs public perceptionTan educates the host on how OpenAI insiders identified compute and data scaling laws long before the public recognized LLMs as more than toy horseless carriages.
Andrew holds their own ▶ 16:07 Warner cites YC's own batch data to counter broad AI tool theoryWarner references David Jim's generalized model theory and immediately counters with concrete batch examples from YC's current cohort to test Tan's investment thesis.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | Andrew as informed peer | Guest teaching | Guest disagreement | Andrew pushing back | Why |
|---|---|---|---|---|---|---|
| Episode Preview: AI Revolution and Modern Startup Velocity | 5 | 3 | 1 | 2 | Warner opens the interview by framing the transformation of early legal-tech startup Casetext before and after the advent of LLMs. Tan explains how historical Web 2.0 SEO models capped out at modest revenues before LLM technology turned them into potential multi-billion dollar opportunities. | |
| Overcoming Early Hallucinations and Discovering AI Scaling Laws | 3 | 6 | 1 | 1 | Tan details the technical limitations of GPT-3 in legal contexts due to hallucination risks, explaining the internal discovery of log-linear scaling laws by OpenAI researchers. Warner largely listens as Tan walks through the technical evolution. | |
| Pioneering Context Engineering and Deterministic Legal Prompts | 4 | 6 | 1 | 1 | Tan outlines how Jake Heller broke down legal queries into deterministic, bite-sized context chunks to establish reliable prompt engineering workflows. Warner summarizes the mechanism effectively. | |
| The Enron Demo and Collapsing Enterprise Sales Cycles | 5 | 5 | 2 | 3 | Warner questions how vertical AI startups can compete against foundational model providers like OpenAI and Google. Tan responds with data on YC batches growing at 10 to 20 percent weekly by targeting massive labor spend in fragmented niches like HVAC rather than standard SaaS seats. | |
| Vertical Specialization, AGI Speculation, and Market Realities | 6 | 4 | 3 | 4 | Warner pushes back against the generalized single-tool vision advocated by Read AI's founder, citing specific newly launched YC companies to show vertical focus. Tan discusses Sam Altman's shifting views on startup moats and lean building. | |
| Vibe Coding, 20x Developer Leverage, and Everyday Quality | 4 | 5 | 2 | 3 | Warner asks whether vibe-coding and AI-assisted apps will create sustainable businesses or merely ephemeral hobby projects. Tan argues that 20x engineer leverage enables software quality to reach unserved everyday verticals, comparing it to Apple's inability to fix calendar bugs. | |
| Generational Talent Shifts and Big Tech Corporate Inertia | 4 | 6 | 2 | 2 | Warner presses on whether spreading software revenue across micro-SaaS companies aligns with YC's unicorn-hunting model. Tan reframes the discussion around talent, noting a 100 percent increase in college-age founders applying while ZIRP-era big tech employees hold onto safe corporate roles. | |
| Y Combinator's Lifelong General Partner Mentorship Model | 5 | 4 | 1 | 2 | Warner highlights YC's historical role in guiding founders through pivots like Jasper. Tan clarifies YC's restructured General Partner model, explaining that partners remain lifelong advisors rather than transient seasonal counselors. | |
| Garry Tan's Iterative Prompt Engineering for Content Creation | 5 | 5 | 1 | 2 | Tan walks step-by-step through his personal metaprompting system for scripting YouTube videos with Gemini and ChatGPT reasoning models. Warner engages on prompting mechanics and convinces Tan to share the exact prompt template with viewers. | |
| Refocusing YC by Discontinuing the Continuity Fund | 6 | 5 | 2 | 5 | Warner probes into Tan's internal restructuring of YC, specifically pressing on why shutting down the Continuity Fund was necessary and why competing with downstream VCs created friction. Tan explains that eliminating fund compartmentalization returned YC to a unified early-stage focus. | |
| Evaluating Opportunities and Design Challenges in Consumer AI | 6 | 4 | 2 | 3 | Tan brings up consumer AI opportunities such as Rosebud AI. Warner immediately validates product-market fit from personal use but provides sharp critique regarding its lack of polished design and voice latency. | |
| The Era of 200x Engineers and Startup Resurgence | 5 | 4 | 1 | 3 | Warner questions whether light AI wrapper apps can transition into defensible long-term enterprises. Tan responds that combined AI tooling turns strong coders into 200x engineers, creating a massive post-earthquake reset across the tech landscape. |
Statements from this episode (18)
Tan: Founders Worry Prematurely About Market Size Ceilings
“I think a lot of founders are worried about that early, but my sense is maybe it's premature worry because embedded in that is also the case text pivot that, you know, they got users and an understanding and a useful corpus of data All of which turned into a t…”
Tan: Scaling laws revealed the path to AGI for OpenAI's leaders
“If you were Greg Brockman or Dario Amadei, ah, at that moment, you started, you know, internally you were talking about the scaling laws, and that the loss function was going down as log linear to the amount of, ah, data and compute you were putting in, and th…”
Tan: OpenAI Was Spun Out of YC Research by Sam Altman
“OpenAI itself was a spin-out from YC research by, ah, Sam Altman.”
Tan: Reliable LLM output requires decomposing tasks into small contextual steps
“At that moment if you chopped it down to a bite-sized chunk, like you gave it some amount of context, That a human being, given the same context and the same prompt, would answer in a certain way. He found that he could, ah, you know, given inputs and outputs,…”
Tan: Casetext founder Jake Heller pioneered early prompt engineering workflows
“I think of Jake a little bit like the first man on the moon.”
Tan: Small YC startups routinely hit $20M revenue in 20 months
“We're seeing routinely YC companies with 10 or 20 people get to 10 or twenty million dollars a year in revenue in 10 or 20 months. And that's like literally never happened before in software.”
Tan: YC startups average 10% weekly revenue growth over 12-week batches
“The average rate at which YC companies grow their revenue during the YC batch, which is a 12 week process, is a 10% per week on average. I mean, some of the batches have been growing 15, 20% a week, but it's been at least 10% a week for more than a year.”
Tan: Anthropic's Claude Code team uses AI to write 95% of codebase
“The Claude Code team apparently writes, 95% of their code is written by Claude, which means very directly that each engineer working on Claude Code themselves is doing the work of 20 people.”
Tan: Founding a startup accelerates your career by 5 to 10 years
“It's better for people to, ah, become founders, learn how to create things for other people, and then either, you know, you manage to get product market fit, and you figure out a moat so that you can be, you know, as big a company as possible. Or even if you d…”
Tan: 18-to-22-year-old founders at YC are up over 100% year-over-year
“Well, even at YC, like, we are seeing the rate of 18 to twenty-two-year-olds at YC is up by more than a hundred percent year on year. The rate of twenty-two-year-olds to twenty-five-year-olds is up about 20%. And then the rate of 25 to thirty-year-olds is actu…”
Tan: ZIRP-era tech workers are clinging to jobs and denying AI
“A lot of those people started their careers actually during Zerp, so they're actually sort of, you know, hanging on for dear life at both startups and Fang, and those are also, like, funny enough, some of the people who are the biggest AI deniers. Like, they j…”
Tan: Y Combinator has 15 equal investing partners selecting startups
“We have 15 equal partners on the investing side who like are in, we basically are in there trying to fish.”
Tan: Gemini 2.5 excels at extracting structural patterns from long video scripts
“One of the things that I did recently, I took the scripts of all of the top videos that I ever made from my YouTube channel, and I just fed it in, and the prompt to Gemini, because it had long context at the time, I think a lot of other people have long contex…”
Tan: Asking AI to iteratively update its own prompt compounds output quality
“Get the script to where I felt really good about it, and then at the last point, I would say, given what we did in this session to improve the prompt, output the next version of the prompt. And so now I've done this about 20 or 30 times, and so now I have a th…”
Tan: YC board abandoned conglomerate model to return to focused accelerator
“When I came back, like, we explicitly decided as a board and as, like, the sort of steering body of what YC was supposed to be, we said, you know what, like, Alphabet is great, but we're gonna go back to being Google.”
Tan: ChatGPT is the most astonishing consumer product launch in history
“ChatGPT itself is actually the best and most astonishing consumer launch of any product in the history of products actually.”