Garry Tan: New Rules for Founders
Time-sensitive content: This talk discusses current AI model capabilities, pricing, and specific short-term predictions about AI's impact on business in the next 2-5 years, including mentions of 2026 and 2028.
Core: Founders today must embrace radical ambition and leverage AI agents to build exponentially better companies, moving beyond traditional SaaS models by pursuing earnest, first-principles approaches to deeply understood problems, while recognizing that AI-powered agency will profoundly transform business and society at a pace moderated by human bureaucracy.
One-line summary
Garry Tan and Anish Acharya agree that founders must radically reimagine their ambition and leverage AI agents to create fundamentally new business models, while recognizing that institutional inertia will moderate the pace of societal transformation—a dynamic that creates both opportunity and responsibility for startup founders.
Participants
| Name | Role | One-line stance |
|---|---|---|
| Anish Acharya | Host | N/A |
| Garry Tan | President and CEO of Y Combinator | Advocates for ambitious, first-principles founders leveraging AI to create profoundly better businesses and societies, emphasizing the transformative power of AI agents and local community engagement |
Arguments
Argument 1 — Founder Psychology & Early Career Mistakes 02:22
- Garry Tan: Founders should be earnest, trusting their unique knowledge and direct experience rather than chasing fads. He shared his regret of leaving web programming for mobile and turning down Palantir because he prioritized "what was hot" or "cool" over what he knew and the smart people involved. (reasoning: His own experience showed that following the crowd—"web is dead, mobile is hot"—led to missed opportunities like Facebook. Prioritizing a promotion over a unique opportunity with brilliant people at Palantir cost him a two to four billion dollar mistake. This highlights the importance of first principles and courage in one's convictions.)
Agreed — Chasing what's "hot" is often the wrong strategy; founders need courage to pursue what they uniquely understand and trust their direct experience.
Argument 2 — Silicon Valley Culture & Innovation 07:11
- Garry Tan: Silicon Valley's true innovation comes from "fringe weirdo punks" and outsiders pursuing "toy-like" ideas that defy conventional wisdom. Y Combinator helps these individuals find their tribe and access the ecosystem. (reasoning: Historical examples like the Homebrew Computer Club and the early personal computer movement demonstrate that foundational ideas often start on the fringes. YC's model of accepting diverse founders, regardless of background, and providing a community fosters this type of unconventional innovation.)
Agreed — The most important innovations emerge from outsiders pursuing seemingly unimportant ideas that become foundational.
"Everything that's awesome in my life is kind of a cult."
Argument 3 — Impact of AI on Founders & Business Models 16:03
- Garry Tan: AI, particularly agentic coding, enables individuals to be exponentially more productive (400x) and necessitates founders to be far more ambitious. Traditional SaaS models are becoming obsolete, requiring new moats based on data or network effects. (reasoning: The ability of AI to automate complex coding and PM tasks means a single founder can achieve what previously required a large team. This changes the competitive landscape, making simple per-seat SaaS unsustainable without deeper competitive advantages, as "code is no longer precious.")
Agreed — The traditional SaaS model focused on per-seat revenue may not be viable in 5-10 years without strong moats like data or network effects.
Founders are now experiencing a fundamental shift: they can be "a multiple of themselves" through AI. The old "iron laws" of business—revenue per employee, per-seat pricing—are breaking down. Companies must evolve beyond simple subscription models to survive the AI transformation.
Argument 4 — AI-Powered Business Operations & 'Skill Files' 22:04
- Garry Tan: AI agents, combined with "skill files" (markdown files + code + tests), can automate nearly every business process within a company, acting as perfect, repeatable "employees" that self-improve. (reasoning: Through personal experimentation with GStack and Gbrain, Garry found that agents can manage engineering, QA, and other tasks, reducing human effort to bottleneck identification. He notes companies achieving 0 to 15M ARR in 4 months with just a few people and hundreds of skill files, demonstrating immense efficiency gains.)
Agreed — "A markdown file is an employee" that performs tasks perfectly and repeatedly once documented.
The implications are profound: founders should "token max," giving agents vast context (e.g., a million tokens) to simulate living in 2028, even if currently expensive. This allows them to identify and eliminate bottlenecks in business processes, enabling a CEO to "live in the future today."
"An org like Microsoft can't. But a startup can. And every startup must."
Argument 5 — AI for Management & Organizational Structure 30:26
- Garry Tan: AI can provide "clairvoyance" to managers and CEOs, allowing them to overcome human cognitive limits (seven plus or minus two things in their head) by analyzing data and conversations across the organization to identify problems and resolve conflicts objectively. (reasoning: He cites Pedro from Brex using OpenClaw to analyze meeting transcripts and network traffic, gaining real-time insights into compliance issues and ground truth two levels down in the organization. This capability can prevent businesses from failing due to management challenges and information silos.)
Agreed — AI agents can transform management by providing real-time intelligence from data and meeting transcripts, enabling leaders to understand ground truth and make decisions more effectively than human-limited managers.
| Management Challenge | Traditional Approach | AI-Enhanced Approach |
|---|---|---|
| Cognitive limits | 7±2 items in working memory | Unlimited context analysis |
| Organizational visibility | 1-2 levels down at most | Real-time insights multiple levels deep |
| Conflict resolution | Based on partial information | Objective analysis of all meeting data |
| Problem identification | Reactive, often too late | Proactive pattern recognition |
Argument 6 — Future of Computing & AI Adoption 41:20
- Garry Tan: The next computer will likely be voice-based, with advanced memory and a benevolent, always-assisting presence. While AI's potential is vast, its societal adoption will be slower than expected due to human and institutional bureaucracy, creating a "white pill" scenario where society has 20 years to adapt. The dropping cost of frontier models will lead to "harness wars" for consumer AI. (reasoning: The current form factor of computers is unlikely to last. Advanced memory, diarization, and context will be crucial for the next generation of AI-powered devices. The inherent slowness of large organizations and governments acts as a "moat," delaying radical change but also providing a buffer for adaptation, which is ultimately a good thing for society.)
Agreed — Bureaucracy and human limitations will slow down AI adoption, giving society approximately 20 years to adapt, which creates structural moats for existing large companies while startups can move faster.
The next computing paradigm will be "all watched over by machines of love and grace"—voice-based interfaces with robust memory, constantly learning user hopes and desires. As compute costs drop, 2027 may see the beginning of "harness wars" for billions of consumers.
Argument 7 — Local Politics & Societal Improvement 45:10
- Garry Tan: Problems in local politics are "human coordination bound," but active engagement from citizens, especially the tech community, can drive significant positive change. Fixing local issues earnestly can improve state and national systems over time. (reasoning: Garry shares his personal motivation and experience in San Francisco politics, where organized citizen action led to positive changes in education and public safety. He believes that tech individuals, who comprise a significant portion of the city's populace, have a responsibility to speak up and hold institutions accountable, acting locally to effect broader change.)
Agreed — Engaging in local politics and community issues is crucial because local problems, if addressed earnestly, can create a functional system that eventually influences state and national levels.
"We may never achieve a utopia, but it is worthy and worth it to attempt."
Agreed
- Earnestness over hype: Founders must trust their unique knowledge and pursue deeply understood problems rather than chasing what's fashionable
- Silicon Valley's fringe culture: The most transformative innovations come from outsiders pursuing "toy-like" ideas that seem unimportant at first
- AI's exponential impact: Agentic AI enables founders to be 400x more productive, fundamentally changing what small teams can accomplish
- Death of simple SaaS: Per-seat revenue models are becoming obsolete without strong data or network effect moats
- Skill files as employees: Markdown files combined with AI agents can automate and perfect business processes
- AI-powered management: Leaders can overcome cognitive limits through AI analysis of organizational data and communications
- Gradual AI adoption: Bureaucratic inertia will moderate the pace of change, giving society time to adapt (approximately 20 years)
- Voice-based future: The next computing paradigm will center on voice interfaces with advanced memory and context
- Local action matters: Tech leaders have a responsibility to engage in local politics to drive broader systemic change
Split
No significant disagreements emerged during this conversation. Anish and Garry were aligned on the fundamental transformations facing founders and society.
Striking moments
- 09:25 — Garry recounts his "two to four billion dollar mistake" of turning down an early opportunity at Palantir, prioritizing a Microsoft promotion over a clear "first principles" opportunity with smart people
- 14:15 — Garry highlights the importance of co-founders, using the analogy of a lone dancer being seen as crazy until a second person joins, transforming it into a party
- 27:10 — Garry states that "a markdown file is an employee," explaining how AI agents can turn documented business processes into perfectly executed, reusable automation
- 34:03 — Garry details how Pedro from Brex uses OpenClaw and "crab trap" to gain "clairvoyance" into his organization, analyzing meeting transcripts and data to understand internal dynamics and conflicts two levels down
- 39:20 — Garry introduces the "white pill" concept, suggesting that the inherent bureaucracy and slowness of human institutions will temper the pace of AI adoption, giving society 20 years to adapt
- 47:00 — Garry explains his personal motivation for engaging in San Francisco politics, connecting it to his own educational opportunities and the need to address issues impacting Asian-American communities
Open questions
- What specific moats beyond data and network effects will sustain SaaS businesses in an AI-dominated landscape?
- How will the "harness wars" for consumer AI unfold as compute costs drop?
- What governance structures will emerge to manage AI agents that have extensive organizational knowledge?
- How can founders balance the push for radical ambition with the realistic constraints of market adoption?
- What new business models will replace per-seat SaaS as AI agents automate traditional software functions?
- How will educational institutions adapt to prepare founders for an AI-native world?
Related concepts
- Y Combinator
- Agentic coding
- AI agents
- First principles thinking
- Silicon Valley culture
- Skill files
- Token context windows
- Network effects
- Data moats
- Voice computing
- Local politics and tech
- Organizational intelligence
- Brex
- Palantir
Key timestamps
- 00:00 — Introduction: Utopia, cults, and startup speed vs. big orgs
- 02:22 — Garry's early career in 2003 Silicon Valley: chasing what's hot vs. earnestness
- 07:11 — Silicon Valley culture: outsiders, fringe ideas, and finding your tribe
- 09:25 — The Palantir mistake: a $2-4 billion lesson in first principles over fads
- 14:15 — YC's role in democratizing Silicon Valley access and building real communities
- 16:03 — AI's impact on founders: exponential ambition and the changing landscape of SaaS
- 22:04 — AI agents and skill files: automating business processes and building companies with markdown as employees
- 30:26 — AI for management: 'Clairvoyance' and overcoming human limitations in organizations
- 39:20 — The 'white pill': why AI adoption will be slower than expected and what it means for society
- 41:20 — The next computer: voice, memory, and the coming 'harness wars' for consumers
- 45:10 — Local politics and human coordination: Garry's engagement in San Francisco and the power of local action