Uncapped with Jack AltmanY Combinator in the Age of AI | Ep. 43
CHAPTERS
How YC’s core value proposition has (mostly) stayed the same since 2006
Jack asks how YC has changed from the earliest batches to today. Jared argues that surprisingly little has changed by design: YC’s original “product” worked, so the organization has tried not to disrupt the fundamentals.
- •Early batches (2006–2008) vs. now: more continuity than outsiders expect
- •YC as a “product” created by Paul Graham that they intentionally preserve
- •Some operational changes over time, but the core experience remains consistent
- •Framing the conversation around YC’s founder-facing value proposition
YC as a transformation engine: community, calibration, and a stamp of approval
The group describes YC less as information and more as a social environment that transforms founders. Garry characterizes it as a place that turns earnest builders into formidable operators; Jack adds the idea of normalization and calibration around startup life.
- •YC as “Disneyland for founder transformation” / “Hogwarts” metaphor
- •The power of being surrounded by serious builders who share a language
- •YC’s credibility signal still matters even with more content online
- •Talent identification: pulling outsiders into the startup ‘vortex’
AI coding tools change the founder archetype: from “great engineer” to “great builder with agency + taste”
Garry explains how tools like Claude Code/Codex compress years of engineering into days, dramatically changing what’s possible for small teams. This shifts what YC looks for—from pure engineering pedigree toward evidence of craft, systems thinking, and product judgment.
- •AI-enabled software velocity: rebuilding a 70k-line product in ~90 hours
- •AGI ‘for code’ as a practical inflection point in late 2024/early 2025
- •Core evaluation traits: agency (belief you can solve problems) and taste (product judgment)
- •Craft signals: edge cases, release completeness, avoiding over-engineering
YC applications now include AI-build evidence: prompting transcripts and ‘game recognize game’
To adapt selection to AI-era building, YC adds the option to submit coding-agent transcripts showing how applicants build features. The partners discuss how prompting style can reveal systems thinking and craftsmanship, similar to how great builders notice details others miss.
- •New application artifact: upload Codex/Claude transcripts for feature-building
- •Security sandboxing to reduce prompt-injection or gaming
- •Evaluating prompting: planning, system design instincts, bug/edge-case handling
- •“Back of the cabinet” craftsmanship analogy; spotting builder quality beyond resumes
MVP quality is rising: faster building raises the bar and increases pivot capacity
Jack notes the bar for product quality is higher than before, even early in a startup’s life. With AI-assisted development, YC expects founders to produce more before interviews and to test/pivot faster during the batch.
- •Advice shift: still ship fast, but expectations for polish/quality are higher
- •Product showcase demos show rising bar every batch
- •AI may enable more pivots within a single batch timeframe
- •Founders can run more experiments quickly without ‘wasting’ years
When to pivot vs. persevere: founder psychology and avoiding ‘random walk’ iteration
Jared describes YC’s role as closer to therapy than a rigid doctrine: the right move depends on founder energy and conviction. A common anti-pattern is launching unrelated ideas hoping the world chooses; instead YC pushes founders to find something they genuinely care about.
- •Pivot advice is situational; partner reads the founder’s excitement and conviction
- •Anti-pattern: no prior on a good idea → try five unrelated things
- •Better approach: dig for a thesis founders care about, then shape it into a startup
- •Traction search should be guided by insight, not lottery-ticket experimentation
What trends YC is seeing: ‘mostly AI,’ with glimmers like prediction markets and stablecoins
The panel says the dominant trend is still AI across the batch, with a few notable pockets. Prediction markets (inspired by Kalshi) and crypto/stablecoins show momentum, often catalyzed by regulatory shifts.
- •AI saturation: most startups now ‘look like AI companies’
- •Smaller trend: prediction markets gaining energy post-regulatory green lights
- •Kalshi as an inspiration model for a new wave of founders
- •Crypto/stablecoins as another non-AI area showing renewed motion
Capital dynamics in the AI era: easier early revenue, bigger Series B’s, and ‘flight to quality’
They discuss why some startups reach $1–2M ARR with tiny teams, yet later rounds are larger than ever. A consolidation toward mega-funds concentrates dollars, making fundraising and scaling patterns feel contradictory across stages.
- •New pattern: hitting meaningful ARR with ‘no hires’ due to tooling leverage
- •Series B rounds appear bigger than ever despite early-stage efficiency
- •Mega-funds concentrating capital across fewer firms/decision-makers
- •‘Capital as a bludgeon’ works in some consumer markets but may be weaker in fast-moving AI
Competition in crowded markets: the default advice is ‘ignore it and execute’
Jack asks how YC advises founders when dozens of startups pursue similar ideas. Partners emphasize that during the batch the key question is whether there’s a glimmer of product-market fit; worrying about competitors often prevents capable teams from launching at all.
- •Batch focus: ‘Is there anything here?’ matters more than competitive mapping
- •Common failure mode: founders quit because incumbents look unbeatable (e.g., Harvey vs. Legora)
- •YC mantra reinforced: ‘Make something people want,’ not ‘build a market map’
- •Competition becomes real only insofar as it blocks customer acquisition
Is SaaS dead—and what’s ‘safe from AI’?: moats shift to systems of record, regulation, marketplaces, and atoms
Garry argues traditional SaaS is vulnerable unless rebuilt with an agentic, top-to-bottom workflow. They debate which businesses are insulated: marketplaces and regulated/touching-money systems feel safer, while integration-heavy SaaS moats are increasingly brittle.
- •SaaS isn’t ‘dead’ if it becomes agentic; legacy approaches are at risk
- •Moats that may persist: marketplaces (Airbnb/DoorDash), systems of record, regulatory workflows
- •Integration moats weaken when code/connectors can be generated quickly
- •Provocative take: Salesforce-like CRM dominance may be disrupted by AI-native entrants
AGI, ASI, and swarm intelligence: what’s next for AI capability leaps
Garry and Jack discuss whether superintelligence arrives as a clear moment or through incremental ‘limited ASI’ systems. They highlight recent examples of multi-agent/swarm behavior and contrast it with the ‘God model’ approach pursued by major labs.
- •Claim: ‘AGI for code’ is here; ASI feels near-term
- •Swarm intelligence as an alternative path to single massive models
- •Multi-agent coordination demos as a validation of swarm research
- •Implications for startups: new capabilities arrive faster than planning cycles
Broadening the founder funnel: campuses, global outreach, and reaching ‘late bloomers’
YC wants more exposure to founders who don’t think YC is for them, so it’s investing in in-person community presence. Efforts include college tours, international trips, and expanding focus to grad students and mid/late-20s professionals, not just undergrads.
- •Problem: YC can feel ‘in the sky’—intimidating and distant to outsiders
- •Boots-on-the-ground recruiting: 30+ campuses, Europe trips, upcoming India travel
- •Expanding beyond undergrads to grad students and older first-time founders
- •Acknowledging trend toward younger founders while noting many iconic YC founders were mid/late 20s
AI anxiety, jobs, and ‘little tech’: managing societal disruption while enabling new entrants
The conversation shifts to how AI affects public trust, worker fear, and inequality dynamics. Garry argues the solution is more entrepreneurship and more competition, supported by policy work that protects startups’ ability to train models and enter markets.
- •Outside SF bubble: more fear and distrust of AI; uneven adoption patterns
- •Real concern: structural unemployment and wage pressure amid rising costs
- •‘Little tech’ advocacy: enabling startups to compete with incumbents (policy/amicus work)
- •Cultural critique: lack of imagination and over-litigation reduce America’s ability to build
California/SF politics and civic capacity: housing, homelessness, safety, and accountability
Garry explains why he became politically active, emphasizing public safety, education quality, and building housing. He cites local leadership examples (e.g., San Jose’s housing and homelessness metrics) and argues progress requires sustained civic engagement and accountability.
- •Motivation: public safety concerns and deteriorating civic outcomes
- •Housing production and homelessness reduction as practical success metrics
- •Education access (e.g., early algebra) as a pathway to opportunity
- •Call for participation: debate openly, scrutinize performance, and use elections to enforce accountability
YC scale strategy in the AI era: more downstream capital, more great founders, and a decentralized ‘pods’ model
They close by discussing venture capital influx and YC’s operating model. YC benefits from plentiful follow-on capital, but the primary bottleneck is finding and inspiring more great founders; structurally YC now runs many small ‘pods’ in parallel, enabling growth like multiple simultaneous early YC batches.
- •More VC capital is mostly positive: YC needs downstream funding for companies
- •Bottleneck focus: increasing the number of great startups by recruiting more founders
- •Operating shift: decentralization—partners pick companies and run pods (mini-YC’s)
- •Scaling vision: parallel ‘2008-style batches’ with many ‘PG-like’ partners; add programs like Fellows to widen the base
