YC AI Startup School 2025
Y Combinator hosted its AI Startup School event in San Francisco, bringing together 2,500 top undergraduate, master’s, and PhD students specializing in AI. The event featured prominent AI founders and industry leaders, including Elon Musk, Satya Nadella, Sam Altman, Andrej Karpathy, Andrew Ng, Fei-Fei Li, Varun Mohan, John Jumper, Aravind Srinivas, Michael Truell, among others. Designed for students and researchers in computer science, machine learning, applied mathematics, and related technical fields, the conference provided valuable perspectives on the trajectory of AI, essential startup strategies, and emerging industry trends. Below are key insights from the event:
Garry Tan — President & CEO, Y Combinator
Garry Tan highlighted the remarkable accessibility of intelligence through APIs, emphasizing that we’re experiencing a pivotal moment where agency matters immensely. Success in tech, he suggested, relies on pragmatic builders who engage actively with users, execute consistently, and exhibit patience for compounding growth.
Sam Altman — CEO, OpenAI
Sam Altman shared fascinating insights on AI’s evolution, suggesting we are currently in a phase of “product overhang,” where AI capabilities advance faster than the development of applications leveraging them. Altman outlined an ambitious vision for GPT-5, emphasizing multimodal capabilities (text, images, video), improved memory, and proactive agent-based applications. He reinforced the importance of contrarian thinking and perseverance, recounting the skepticism surrounding GPT-1, highlighting how persistence through uncertainty eventually leads to groundbreaking achievements.
Elon Musk — CEO, Tesla & SpaceX
Elon Musk emphasized practical utility over grand aspirations and the importance of truth-seeking cultures within organizations. He shared that effective AI startups require three foundational elements: computational power, unique data, and talented people. Musk’s futuristic outlook included predictions of humanoid robots surpassing human populations within a century, stressing humanity’s responsibility to become a multi-planetary species.
Satya Nadella — CEO, Microsoft
Satya Nadella argued that the true measure of AI should be its economic impact rather than achieving anthropomorphic intelligence. He described AI as a transformative tool reshaping job roles rather than eliminating them, envisioning a future where software engineering transitions into roles like Forward Deployed Software Engineers. Nadella highlighted privacy, security, and sovereignty as critical considerations in AI adoption and deployment.
Aravind Srinivas — CEO & Co-Founder, Perplexity AI
Aravind Srinivas offered insights from building Perplexity, emphasizing the necessity of resilience, continual experimentation, and adaptation to competition. His practical advice included focusing deeply on specific domains (like browsers and search) rather than competing broadly, noting the significant advantages startups have when tackling specialized tasks.
Fei-Fei Li — Stanford Researcher
Fei-Fei Li provided a compelling narrative on the integration of AI and human cognition, emphasizing that language alone is insufficient for AGI. Highlighting the value of spatial intelligence, Li advocated for innovative research driven by curiosity and intellectual fearlessness, reminding aspiring entrepreneurs that startup environments demand practicality over pure exploration.
Andrej Karpathy — Former Director of AI, Tesla
Karpathy introduced the concept of “Software 3.0,” describing a progression from traditional coding to neural network training, and now prompt-based interactions. He equated modern AI models to infrastructure like electricity, stressing their transformative potential across multiple industries, despite existing limitations such as memory gaps and reliability concerns.
Andrew Ng — AI Educator and Founder, DeepLearning.AI
Andrew Ng focused on execution speed as the strongest predictor of startup success, stressing that the most significant AI opportunities currently lie at the application level. He highlighted the shift toward product management as a critical role in AI-driven enterprises and recommended leveraging iterative development and feedback loops to rapidly refine products.
Chelsea Finn — Co-Founder, Physical Intelligence
Chelsea Finn discussed the future of robotics, advocating for generalist robots supported by robust Vision Language Models (VLM). Finn emphasized the importance of diverse, real-world data in training versatile robots, predicting that general-purpose robotics would soon outperform narrowly designed alternatives.
Jared Kaplan — Co-Founder, Anthropic
Jared Kaplan presented compelling data demonstrating rapid advances in AI capabilities, particularly in complex, long-chain tasks. He advised startups to anticipate future AI advancements by building products initially beyond current capabilities, thus aligning product maturity with future AI developments.
François Chollet — CEO, Ndea and Creator of Keras
François Chollet provided a thoughtful critique of current scaling approaches to AGI, arguing that genuine intelligence requires adaptability and innovation, not merely automation. He advocated for AI systems capable of discrete, combinational reasoning rather than purely continuous pattern recognition.
Key Takeaways:
AI is shifting from pure research to highly practical applications.
Economic impact and practical utility are paramount for AI advancements.
Startups should focus on specialized niches and rapid iteration rather than broad competition.
Future AI development will heavily involve multimodal capabilities, memory integration, and adaptable interfaces.
Human roles in tech will evolve significantly rather than disappear, emphasizing roles like product management and system architecture.
AI is not just an evolving technology but a transformative force reshaping the very nature of innovation, business strategy, and human-computer interaction. It’s indeed an exciting time to be part of this industry.