Chronological feed of everything captured from Y Combinator.
OpenAI's success stemmed from a startup-like culture emphasizing scaling laws over academic approaches, validated through projects like Dota 2, robot hand Rubik's Cube, and GPT evolution via next-token prediction at scale. Pre-training faces data walls, but reasoning and test-time compute enable S-curve shifts, boosting reliability for agents and innovators via longer chains of thought. Distillation allows efficient deployment of frontier capabilities; future bottlenecks shift to robotics and UI integration for real-world impact, with optimism for new human roles as geniuses or AI managers.
AI has elevated YC startup benchmarks, with entire batches averaging 10% week-over-week growth and companies achieving $12M ARR in 12 months—previously top-1% feats. High enterprise demand for reliable AI agents enables technical founders to secure massive contracts via superior prompting, evals, and rapid iteration, bypassing heavy sales. This force-multiplies founder agency, expands viable business universes, and shifts hiring/pricing toward leverage and usage-based models amid model commoditization.
Traditional LLM scaling laws, balancing parameters, data, and compute, drove consistent performance gains until recent plateaus from data limits and diminishing returns. OpenAI's o1 and o3 models introduce a new paradigm: scaling test-time compute via extended chain-of-thought reasoning, yielding massive benchmark leaps without larger pre-training. This shift promises sustained intelligence scaling, potentially toward AGI, and extends to other modalities like vision and robotics.
Y Combinator partners advise founders to integrate LLMs into startups—either at the core or for internal efficiency—without pivoting superficially to trendy AI ideas, as fundamental startup execution remains unchanged. They compare the current AI wave to past shifts like cloud and mobile, urging founders to embed in San Francisco's ecosystem for rapid idea generation and insights via proximity to experts. Promising applications include automating specialized tasks like UI localization, security engineering, and healthcare admin workflows, often by observing repetitive jobs firsthand.
Parker Conrad's Rippling succeeds by building a unified platform that interconnects HR, payroll, IT, and other functions around a central employee data layer, enabling seamless automation that point solutions cannot match. This "compound software" approach captures more value per user, invests deeply in cross-cutting R&D like analytics and workflows, and addresses core organizational coordination problems. AI enhances this by expanding managerial context windows, flattening hierarchies to run large firms like small ones through data ingestion and anomaly detection in performance and sales.
Rul, founder of Zip (Series D at $2.2B valuation), contrasts his first venture Flight Car—an operationally intensive, low-margin airport car-sharing business that nearly failed financially—with his high-margin B2B SaaS procurement platform. Key shifts include prioritizing scrappy validation over perfection, working at top companies like Airbnb and YC to learn scalable operations and incentives, and proving product-market fit via cold outbound sales to 10 strangers before scaling. Second-time founding mindset rejects external perceptions (investors, press, team optics) in favor of truth-seeking, disproving assumptions, and focusing board time on business breakdowns rather than successes.
Y Combinator podcasters predict AI securing additional Nobel Prizes in 2025, building on 2024 wins in physics and chemistry, due to accelerating scientific discoveries. Stablecoins will drive mainstream crypto payments as wallets with hundreds of millions of users provide supply-side liquidity, enabling easy merchant adoption to solve marketplace chicken-egg dynamics. Real-time AI Zoom avatars with natural latency and lip-syncing will emerge, evolving from 2024's voice AI calls; Dogecoin may rise if government efficiency cuts (Doge) lower interest rates, with AI acting as a deflationary counter to inflation.
David Lieb's journey began with Bump, a viral contact-sharing app from a YC side project that hit #2 on the App Store but failed due to low-frequency, low-value usage, leading to classic startup errors like over-hiring and premature scaling. User interviews revealed a photo-sharing pivot opportunity, evolving through failed apps like Flock into Photo Roll, sold to Google where internal defiance birthed Google Photos—a billion-user product in under 4 years via AI-driven organization. Leukemia diagnosis amid success prompted leaving Google for YC, emphasizing total commitment, user data over feedback, risk-taking beyond norms, and high-frequency/high-value product design.
2024 marked a breakout year for AI startups in YC, with summer and fall batches averaging 3x revenue growth—equating to Paul Graham's 10% weekly target—driven by rapid enterprise pilot-to-revenue conversion and vertical AI applications. Open-source models like Llama eliminated foundation model monopolies, enabling model-agnostic architectures, multi-model orchestration, and post-training specialization without massive capital. Trends include agentic workflows, voice AI verticals, AI coding tools boosting developer productivity 10x, and early robotics software on commodity hardware, projecting 1,500+ annual $100M ARR companies.
Founders should prioritize building a "minimum evolvable product" rather than just a minimum viable product. This approach focuses on finding and intensely studying early adopters who will shape the product's evolution. The strategy involves targeted outreach, charging real money for sharper feedback, and continuous experimentation to adapt to early user needs, recognizing that initial product design is path-dependent on these early interactions.
Claude Code prioritizes building for future LLM capabilities, anticipating rapid model advancements. This foresight led to its core design principles, such as rapid iteration, a focus on latent demand, and a lightweight, terminal-based interface. The product's success highlights the importance of adaptability and a willingness to discard temporary scaffolding as models evolve. This philosophy has led to remarkable productivity gains at Anthropic.
The emergence of AI agents is creating a new economy where these agents, rather than humans, will increasingly drive the selection and adoption of developer tools and other products/services. This shift necessitates optimizing developer documentation and product design for agent interaction and presents new opportunities for startups to build agent-native infrastructure and tools. The rapid growth of platforms like Moltbook, where AI agents interact autonomously, showcases a tangible example of this evolving ecosystem and the potential for "swarm intelligence." While still nascent, this agent-driven paradigm is poised to extend beyond dev tools, influencing various aspects of daily life as agents become significant economic actors.
The "20x company" paradigm describes startups that leverage extensive internal AI automation across all functions to achieve disproportionate growth and efficiency. This approach enables small teams to outperform larger incumbents by dramatically enhancing individual employee output and delaying hiring for conventional roles. Key strategies include AI teammates, unified AI-powered data sources, and custom workflow agents, transforming startup operational models.
Y Combinator advocates for a new "AI native agency" model, moving away from traditional software sales to a service-based approach. This model leverages AI tools to deliver finished products and outcomes, rather than selling access to the tools themselves. This shifts agency economics by reducing the need for proportional headcount scaling, allowing for software-like margins and rapid expansion.
Emergent, a Y Combinator-backed company, is democratizing software development by enabling non-technical users to build production-ready applications using AI agents. This platform allows individuals and small businesses to rapidly create customized software solutions, fostering a new era of personalized applications and reducing reliance on traditional SaaS models. With 7 million apps built in 8 months, Emergent demonstrates significant market traction and is expanding the definition of who can be a software creator.
Y Combinator experienced organic growth driven by word-of-mouth referrals from successful alumni and the essays of Paul Graham, rather than deliberate marketing. The organization faced significant stress from public perception issues, "wokeness," and managing Hacker News, which overshadowed the internal challenges of supporting startups. Early on, YC prioritized supporting founders by providing crucial advice, such as focusing on specific markets, and refining their Demo Day pitches to highlight key value propositions, ultimately fostering a strong community.
Solugen has developed "chematic processing," a novel method fusing biology and chemistry to produce industrial chemicals. This approach utilizes enzymes and metal catalysts to achieve high reaction yields and offers a cleaner, safer, and more environmentally friendly alternative to traditional fossil fuel-based chemical manufacturing. Their strategy emphasizes modular, scalable production units and a strong customer-centric development process.