No PriorsNo Priors Ep 66 | With Y Combinator President and CEO Garry Tan
CHAPTERS
Garry Tan’s path to YC: early YC as a “punk club” for startup outsiders
Elad and Sarah introduce Garry Tan and set up the conversation as a collaboration episode. Garry reflects on discovering YC in 2008, when it felt more like a subculture than an institution, powered by Paul Graham’s essays and internet-native founder communities.
- •Garry’s roles: founder (Posterous), investor (Initialized), now YC President/CEO
- •Early YC’s identity as an internet-driven subculture before mainstream dominance
- •Paul Graham’s essays as the “on-ramp” for aspiring founders
- •How founder education has expanded from essays to podcasts and YouTube
- •Starting a company becoming more culturally accessible over time
Inside the 2008 YC era: tiny batches, enduring relationships, and long-run reputation
Garry and Elad compare the early YC era to today’s scale, noting how small the batches were and how deeply founders bonded. They emphasize Silicon Valley’s long time horizons and how non-zero-sum behavior compounds through networks.
- •2008 batch size (~25 companies) versus today’s many group partners and hundreds of companies
- •Examples of batch peers and long-term trajectories (Twitter acquisitions, Vicarious/Google)
- •Friendships and collaboration formed “over pizza and beers” endure for decades
- •Non-zero-sum behavior and reputation matter enormously over long horizons
- •Founder ecosystems as compounding relationship networks
Posterous and the first social wave: pattern-matching tech shifts (and missing ‘Instagram’ moments)
Sarah asks about Garry’s early investing and his founder years at Posterous, prompting a look back at the chaotic early social media period. The group draws parallels between the skepticism around early social networks and today’s AI cycle.
- •Posterous’s near-miss at becoming an Instagram-like outcome (in hindsight)
- •Early social networks: many failures that fueled skepticism about the whole category
- •How Facebook’s culture felt unusually intense and “missionary” early on
- •Why “winner” predictions were wrong (Friendster/MySpace → Facebook)
- •Parallels between social/mobile platform shifts and today’s AI transition
Technical founders and “firsthand” truth: what YC selects for (still)
The conversation shifts to what YC looks for in founders, with Garry arguing the core criteria haven’t changed. He emphasizes technical strength, clarity of communication, and deeply customer-grounded insight over generic, hand-wavy pitches.
- •YC’s preference for highly technical founders who communicate clearly
- •Anti-pattern: broad ‘does everything’ ideas without a wedge or user love
- •First-principles approach: talk to customers/users directly
- •Information hierarchy: customers (firsthand) > founders (secondhand) > investors (thirdhand)
- •Focus on a narrow initial user set (thin edge of the wedge)
AI in YC’s pipeline: ‘wrappers,’ IQ levels, and why revenue is exploding fast
Elad frames YC as a ‘founder voting machine’ on what’s interesting in AI, and Garry breaks down what he’s seeing across batches. They argue that “wrappers” can be real businesses, while model capability and workflow engineering drive unusually rapid early revenue growth.
- •~70% of YC companies are AI-related in the current period
- •Two-thirds of AI startups are application-layer implementations (often called ‘wrappers’)
- •Why “wrapper” is a misleading critique (analogy to cloud/SaaS ‘MySQL wrappers’)
- •Models as ‘eager intern’/~85 IQ today—powerful but constrained and supervised
- •Batch-wide revenue growth signals: early teams scaling ARR dramatically in months
From prompts to products: time-and-motion workflow mapping and the GPT-4 inflection
Garry uses the Casetext example to illustrate how LLMs became practical when hallucinations dropped and workflows could be operationalized. Sarah connects this to a broader pattern: pairing model intuition with deep domain process knowledge to build durable products like Harvey-style systems.
- •GPT-4 as a practical turning point for monetizable, lower-hallucination workflows
- •Casetext’s approach: mapping lawyer tasks into prompts, scoring, and testing
- •‘Intelligence on tap’ becomes usable when tied to measurable workflow steps
- •The winning team pattern: technical model operator + domain time-and-motion expert
- •Large fraction of startups now doing the unglamorous work of evaluation and reliability
Application vs infrastructure: tooling sprawl, open-source commoditization, and stack dynamics
The hosts dig into where founders are building—primarily on top of frontier models, with a growing slice focused on infrastructure. Garry highlights a recurring market pattern: heavily funded closed-source products get followed by scrappy open-source alternatives that commoditize capabilities.
- •Rough split: majority application-layer, increasing minority infrastructure/tooling
- •Tooling categories: testing, prompt collaboration, evaluation, and dev workflows
- •Proliferation akin to LangChain/LangFuse ecosystems
- •Closed-source, heavily funded entrants vs open-source commoditizers
- •Multiple layers evolving in parallel, reshaping startup defensibility
Can scrappy startups train foundation models? Milestone-driven execution (the Cruise analogy)
Sarah asks whether YC expects teams to pre-train models given compute-heavy economics. Garry argues some will, but the playbook remains milestone-driven: build a compelling demo that attracts capital and talent—similar to Cruise’s early highway-only autonomy demonstration.
- •Some YC companies are building foundational models (bio, robotics, etc.)
- •Compute costs force founders to plan backward from large funding needs
- •The YC approach: identify near-term milestones that prove the trajectory
- •Cruise as precedent: a constrained demo that unlocked resources and momentum
- •A small initial check can be enough to reach the proof point that changes everything
YC’s “shipping culture”: group office hours, social pressure, and compounding speed
Elad asks where YC’s execution intensity comes from, and Garry traces it to multiple cultural inputs—from Paul Graham’s vision-setting to Paul Buchheit’s group office hours. The result is structured social accountability that forces weekly progress and rewards sustained speed over time.
- •PG’s role: helping founders see the long-term scale of the wedge
- •Paul Buchheit’s group office hours as an accountability mechanism
- •Mild ‘co-opetition’: competing with friends to avoid showing up unprepared
- •Weekly goals, tangible progress, and customer contact as default expectations
- •Running fast for a long time turns into compounding outcomes
Startups vs elephants: why incumbents don’t always stomp—and the dangers of raising too much
Sarah explores how to choose companies that won’t be crushed by big tech, and Garry argues early-stage is about plausibility and closeness to reality. They discuss examples like Scale and why large incumbents can become less nimble; then pivot to how excess capital can make companies unable to make necessary cuts and course-correct.
- •‘Mice at the feet of the elephant’: evaluating survival against incumbents
- •Scale as an example where big players ‘could have’ built it but didn’t
- •Big tech moats can paradoxically reduce their urgency and execution speed
- •Lesson learned: raising too much money can distort behavior and decision-making
- •Downsizing and painful resets become harder when organizations grow too large
YC founder demographics and fundraising shifts: age cycles, market openness, and bigger checks
The discussion turns to how YC has changed over time—founder age, what gets funded, and how the YC deal size influences who applies and what gets built. Garry links founder demographics to market openness: in fast-moving booms like AI, younger founders can compete; in harder, domain-heavy markets, founders skew older.
- •Average founder age around Garry’s YC age (~27), with recent shifts younger
- •AI boom pulls in 19–21-year-old founders who don’t want to miss the wave
- •When markets are ‘closed,’ domain expertise matters more and founders skew older
- •YC deal increases historically preceded standout cohorts (Instacart/Coinbase/DoorDash)
- •Today’s median fundraising amounts are higher due to brand compounding and founder quality
San Francisco and civic renewal: in-person density, public schools, and speaking up with nuance
Garry argues SF’s in-person conversations accelerate AI progress by bringing researchers, tool builders, and deployers together locally. The conversation then broadens into civic engagement: effective government, public education (especially math), and the importance of speaking openly and voting to restore institutional excellence.
- •SF’s unique ‘density advantage’ for high-leverage AI conversations and collaboration
- •YC’s aspirational role as a beacon for builders and creators
- •Civic agenda: effective government and strong public schools as non-negotiables
- •Math/algebra access as a crucial ladder for opportunity—especially for families without wealth
- •Call to action: discuss issues, educate yourself, vote, and re-enable nuanced discourse
