Y Combinator
Showing 1–15 of 860 transcripts.
- 49 min
Why Everyone Is About To Have Their Own Database
Supabase has accelerated its transition to an agent-first platform, now supporting 13 million users and generating over 70% of its activity through autonomous AI agents rather than human developers. By re-architecting its infrastructure for sandboxes, integrating error-correction logic for autonomous systems, and acquiring Terso to offer a lightweight SQLite tier, the company enables a new demographic of non-technical "builders" to create software via natural language prompts. Co-founder Paul Copplestone projects that this shift will democratize entrepreneurship by 2027, allowing individuals to build and customize applications without traditional development teams while maintaining the platform's commitment to open-source transparency and data ownership.
- 1h 6m
The New Full-Stack Designer
Marvin Schwaibold, Aaron Epstein
Marvin Schweibold and Shopify's Product Design Studio have transitioned their workflow from static design to AI-driven development environments, enabling designers to generate functional code and simulate complex systems without relying on traditional engineering handoffs. By acting as editors within this new paradigm, the team successfully executed parallel initiatives to visually polish the merchant admin and shift navigation to conversational commands, while case studies involving shader physics and carousel animations demonstrated the ability to decompile high-fidelity motion design through targeted AI prompting. Concurrently, the presentation critiqued six contemporary websites to highlight recurring user experience pitfalls in AI tool interfaces, offering specific architectural and interaction recommendations to improve clarity, conversion, and usability for NStack, SkillSync, InstaPlay, Mintlify, Nullify, and Instinct.
- 1h 51m
Why Inference Matters 12x More Than Training | YC Paper Club
Dan Fu, Jeff Morgan, Joe Fioti, Lianmin Zheng, Francois
DeepSeek, Together AI, and other industry leaders are transforming AI economics by shifting focus from training to inference optimization, where compute costs now dominate due to high token volumes and long model lifespans. Through innovations in kernel compilers, disaggregated system architectures, and model-specific techniques like Mixture of Experts and quantization, providers are achieving gross margins near 80% while delivering open models at a fraction of the cost of proprietary alternatives. Strategic advancements in runtime scheduling, KV cache retention, and hardware-at-compiler levels address critical bottlenecks in decode throughput, enabling robust, low-latency inference across hybrid local-cloud environments.
- 47 min
A Different Approach To Launch Videos
Founders are advised to replace high-budget product launches with a strategy of constant, authentic mini-launches on algorithmic platforms like Instagram and X, as demonstrated by HeyClicky's success in generating millions of views and thousands of downloads through raw video content. This approach prioritizes showing the build process over polished personas, utilizing a dedicated filmmaker to produce scripted videos that focus on problem-solution hooks rather than traditional press outreach. By adopting this "show your work" methodology, creators can build a loyal community and a sustainable competitive moat before their product features can be easily cloned.
- 48 min
Magic Patterns CEO Reviews Startup Websites
Alex Danilowicz, Aaron Epstein
Launched in October 2023 by Y Combinator alumni Alex Danilovich and his co-founder, Magic Patterns is an AI design tool that converts text prompts and code into production-ready prototypes to assist non-designers. The platform distinguishes itself through its Canvas OS and Model Context Protocol integration, enabling direct connections to coding environments like Cursor while supporting model-agnostic workflows. By combining rapid Figma-to-code translation with a "code as medium" roadmap for complex 3D interactions, the tool targets product managers and founders seeking to bypass traditional design bottlenecks.
- 1h 37m
AI Safety Fact vs. Fiction | YC Paper Club
Clark Barrett, Mykel Kochenderfer, Francois
The YC AI Safety Club convened to reframe artificial intelligence discourse from sensationalist extinction scenarios to rigorous engineering standards advocated by figures like François, Yann LeCun, and Jensen Huang. Speakers presented concrete mitigation strategies, including Clark Barrett's automated formal verification of cryptographic code and Michael Kochenderfer's adaptive stress testing methods modeled after aerospace safety protocols. The event concluded by proposing a market-driven regulatory framework where independent reporting agencies and zero-trust security principles replace ineffective hardware bans while prioritizing cyber defense against nation-state threats.
- 43 min
How Jeetu Patel Runs Cisco Like the World’s Largest Startup
Cisco CEO Jitu Patel is repositioning the company as a full-stack infrastructure provider for the AI era, driving a rapid business expansion that reached $9 billion in orders by the recent fiscal year. To support this growth and address unprecedented network demands from AI agents, Patel enforces a mandatory "go all-in" AI adoption policy while restructuring leadership to blend seasoned operators with external experts and startup founders. This "wartime" operational strategy aims to secure the supply chain and security of a rapidly scaling market, positioning Cisco to capture the next phase of artificial intelligence integration across silicon, photonics, and software.
- 1h 21m
What If We Stopped Using GPUs? | YC Paper Club
This event examines the stagnation of FLOPS-based efficiency in current AI hardware and proposes alternative architectures that align with biological constraints, such as feed-forward pathways and the elimination of backpropagation's weight transport problem. Speakers detail emerging solutions including zero-order optimization techniques like SPSA combined with the Sharded Optimization Mixture of Assemblies strategy, alongside hardware innovations in optical computing and neuromorphic systems that prioritize energy scaling and memory-compute co-location. The discussion culminates in a case study on biological computing, specifically the collaboration with Cortical Labs to train human brain cell cultures to play *Doom*, illustrating a novel paradigm where intelligent behavior emerges from entropy minimization in a distributed, non-digital substrate.
- 30 min
New LLMs Are Unlocking Robot-Use Agents
Francois, Hanmei, Vincent, Jay
Frontier researchers and startups like Waddle Labs are advancing a paradigm shift toward general-purpose "robot use agents" that leverage coding-based policies and cross-modal data to perform complex physical tasks without specialized fine-tuning. By applying the "Bitter Lesson" principle to consolidate diverse datasets, these systems utilize in-context learning and memory consolidation architectures to rapidly adapt to new environments, with current demonstrations successfully executing multi-step manipulation challenges. Industry projections indicate that as latency decreases and internal world representations converge, economically viable, general-purpose robots capable of performing competent human-like tasks will emerge within two years.
- 36 min
Startups Are Moving From Bits to Atoms
Recent YC batches have witnessed a hard tech resurgence, with the physical sector share tripling to 20% alongside a historic 1-in-6 PhD founder rate driven by AI acceleration and macroeconomic shifts in defense and space. Concurrently, the software landscape has pivoted toward revenue-generating agentic workflows and solo founders, as AI tools enable single individuals to execute complex full-stack operations and experienced founders to manage autonomous coding teams. These trends collectively signal a fundamental shift where advanced physical intelligence, specialized silicon, and autonomous agents are redefining startup viability and scaling strategies.
- 13 min
How To Get Better At Outbound Sales
Founder-led sales strategies prioritize executing at least 100 manual outreach attempts to validate messaging and targeting before attempting automation, ensuring high conversion rates through rigorous prospect research and job title verification. This approach relies on constructing problem-centric, peer-to-peer communication while leveraging customer feedback to build credibility and optimize LinkedIn profiles for maximum trust. Successful execution requires strict operational discipline, including daily time-blocking for consistent volume and a systematic debugging framework that iterates through deliverability, content, and market fit until a viable reply rate of two to three per hundred is achieved.
- 1h 0m
Self-Improving Harnesses, Local Personal AI And YC's Agent For Work | YC Paper Club
Seth Karten, Jon Saad-Falcon, Josh France, Regan Bell
YC Harness Club researchers demonstrate that "self-improving harnesses" capable of modifying their own code and memory have driven Arc AGI success rates from 30% to 100%, surpassing standard prompt engineering by leveraging recursive learning architectures. Presentations by Seth and John Sad Falcone detail specific implementations like Prime Agent's multi-tier context management and OpenJarvis's local on-device stack, which collectively achieve frontier-level performance while reducing costs by 800x and enabling autonomous long-horizon research. Furthermore, the QM project establishes a centralized cloud-hosted framework with dynamic sandboxing and fine-grained permissions, addressing critical challenges in system-wide coordination and data security for scalable multi-agent deployment.
- 57 min
Open Models Are Collapsing The Cost Of AI
Driven by cost efficiency and customization needs, enterprises are increasingly adopting open AI models, with major corporations like AT&T shifting significant token consumption away from proprietary systems toward flexible, locally deployable solutions. Ollama has emerged as a critical orchestration layer for this ecosystem, supporting 85% of Fortune 500 companies and facilitating a hybrid deployment strategy that balances local hardware capabilities with cloud-based inference for complex tasks. This market shift positions open models to handle the majority of daily enterprise workflows, reducing reliance on expensive frontier closed models while navigating geopolitical security concerns through rigorous supply chain verification and secure hosting environments.
- 21 min
Paul Graham On Startups, Ambition, and Great Founders
Paul Graham delivered his 47th Y Combinator batch talk, highlighting a shift toward high-seriousness ventures in logistics and cancer research while defining founder ambition through the "formidable" archetype of those who consistently achieve their goals. Graham analyzed the current AI landscape, noting its uneven "jagged frontier" and the economic reality that token costs now eclipse salaries despite rapid price declines. He concluded that Y Combinator's ecosystem remains vital for peer collaboration and that the next trillion-dollar company will emerge not from a specific idea but from formidable founders capable of adapting to shifting market conditions.
- 15 min
The World’s Largest Electric Aircraft Just Flew
Gustaf Alströmer, Anders Forslund, Clara
Heart Aerospace successfully flight-tested the world's largest electric airplane, a 100-foot wingspan aircraft capable of carrying 36 passengers on short-haul routes with minimal operating costs. Led by CEO Anders Forslund and co-founder Clara, the Los Angeles-based team utilized a hybrid configuration to achieve a 500-mile range while significantly reducing maintenance and noise through the replacement of traditional jet engines with electric motors. This clean-sheet design, which marks the first new American airliner in 18 years, aims to replace inefficient regional jets by leveraging existing airport infrastructure and capitalizing on current advancements in battery technology.