Latest Interviews

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.

Max Junestrand: You Need The Willingness To Learn Faster Than Anyone Else

Max Junestrand, Gustav Omstrammer

Legora has scaled from three Swedish engineers to over 750 employees, generating $100 million in annual recurring revenue while deploying an agentic AI operating system that now serves three percent of the world's lawyers. Following a strategic pivot after a $35 million funding round revealed revenue lagged behind interest costs, the company shifted focus from growth velocity to product-market fit by freezing sales and transitioning to proactive, autonomous agents. Founded by Max, August, and Sigge, the firm differentiates itself through a high-intensity culture designed to overcome Scandinavian humility, recently securing talent at a 23-year-old salesperson who generated over $10 million without prior industry experience.

Going In Deep On Data | YC Paper Club

Francois Chaubard, Vincent Sunn Chen, Volo Kuleshov, Shayne Longpre

Industry experts from Snorkel, Inception Labs, and MIT articulate a paradigm shift where data has evolved from a passive resource into a $100 billion sector requiring complex product craftsmanship and continuous expert supervision. Presentations detailed breakthroughs such as Snorkel's Senior SWE-bench with validation agents, Inception's Mercury 2 diffusion model achieving 1,000+ tokens per second, and novel cross-lingual scaling laws that optimize low-resource language performance. These developments collectively redefine the production frontier by demonstrating that architectural advances alone are insufficient without scalable data programming and domain-specific synthetic environments to resolve real-world distribution gaps.

Susan Kare: How I Designed the Original Apple Mac Icons

Susan Kare

Iconographer Susan Kerr recounts her pivotal role in shaping early Macintosh interfaces, from designing the "Chicago" font to creating iconic symbols like the "Happy Mac" boot screen under strict pixel constraints. Drawing on her experiences at Apple, Microsoft, and Facebook, she advocates for minimalist design principles that prioritize meaningful metaphors over product-specific imagery to ensure universality and longevity. Her career highlights how simple, salient details and strict limitations can foster creativity, ultimately defining the visual language of modern computing interfaces.

Chelsea Finn: This is the State of the Art in Robotics

Chelsea Finn

Physical Intelligence has advanced its PI-07 foundation model to achieve general-purpose physical AI capable of zero-shot generalization across diverse appliances and robots without task-specific retraining. By integrating multi-timescale memory and optimized reinforcement learning strategies, the organization significantly improves task throughput and reliability for complex manipulation scenarios like kitchen workflows. Currently deployed by YC-backed companies for real-world applications, these models represent a shift toward "out-of-the-box" generalist agents that can operate autonomously in unstructured environments.

Why Robotics Still Isn't Solved - But Could Be Soon | YC Paper Club

Francois Chaubard, Marcel Torne, Milan Ganai, Tyler Ga Wei Lum, Niko West, Bill Jiao, Guanming Wang, Shyam Gollakota

Recent industry presentations address critical scaling barriers in robotics, including Sim-to-Real gaps, deformable object handling, and the lack of sensory feedback by introducing specialized architectures like Multi-Scale Embodied Memory and self-supervised Chain of Thought frameworks. Concurrently, a strategic shift toward "Robotics Application Companies" emphasizes proprietary data collection and teleoperation over generalist models, while new World Action Model optimizations reduce inference latency to enable real-time physical AI deployment. These technical and business advancements collectively aim to transition robotics from isolated demos to mass-market, long-horizon task execution.

Max Hodak: Average Is Not Good Enough

Max Hodak

Max Hodak, CEO of deep-tech startup Science, details the company's solar-powered retinal prosthesis that successfully enabled a patient to read a 300-page novel following major clinical trials. The discussion outlines Science's operational strategy, which includes the proprietary Helix purchasing platform to accelerate hardware procurement and an Eigenreviews performance system utilizing eigenvector centrality for continuous employee evaluation. Hodak further argues that success in brain-computer interfaces relies on high-speed iteration, mission-driven leadership, and building custom internal tools rather than relying on generic enterprise software.

How To Design In The Agent Era

Stephen Haney, Aaron Epstein

Led by Stephen Haney, Paper is an AI-native design platform currently ranking as the third most-used tool behind Figma and Sketch, utilizing a native HTML and CSS rendering engine to facilitate precise agent collaboration. The application distinguishes itself by treating code as the single source of truth, offering features like direct React component handoff and an MCP server that allows autonomous agents to manipulate local repositories and design canvases. While the team prioritizes human curation to maintain design quality against common AI aesthetic pitfalls, the platform is rapidly evolving into a visual coordination layer for the broader agent ecosystem by integrating tools like Paper Shaders and multi-model image generation.

Waymo Co-CEO Dmitri Dolgov: The Demo Is Only 1% Of The Work

Dmitri Dolgov

Waymo has advanced its fully autonomous fleet to operate 500 weekly trips across 15 U.S. cities, achieving a safety record 17 times better than human drivers through a multimodal sensor architecture and a foundation model utilizing both fast geometric reactions and slow semantic reasoning. The company addresses the unique challenges of physical AI by integrating real-world data into a generative simulation ecosystem that creates rare edge cases for training, while its "Safety and Readiness Framework" rigorously validates performance to ensure public trust and regulatory compliance. Looking ahead, this structural augmentation and flywheel of agent, simulator, and critic data positions Waymo to expand its technology beyond personal vehicles into trucking and broader physical applications.

Kernels and Chips: Cursor, NVIDIA, and Meta Researchers on GPU Performance | YC Paper Club

Stuart Sul, John, Francois Chaubard, Jon Saad-Falcon, Mark Saroufim, Misha Smelyanskiy, Brennan Shacklett

The event synthesizes critical industry shifts toward specialized ASICs and disaggregated architectures that separate training and inference workloads to maximize intelligence per watt. Technical deep dives highlight breakthroughs in multi-GPU kernel optimization and local inference, demonstrating that consumer accelerators can now deliver nearly 90% of frontier model utility while reducing energy costs by 70%. These innovations collectively address systemic bottlenecks in network communication and latency, establishing a roadmap for distributed, cost-efficient AI ecosystems that minimize reliance on centralized cloud resources.

Blake Scholl: How 50 People Built a Supersonic Jet

Blake Scholl

Founded by Blake Scholl in 2014, Boom Supersonic successfully developed the XB-1, the first privately built jet to break the sound barrier, using a lean team and software-driven engineering to overcome decades of aviation stagnation. This historic flight directly catalyzed a shift in US federal policy, prompting an Executive Order that legalized supersonic flight over land and clearing the path for the upcoming Overture airliner. By vertically integrating manufacturing and creating revenue through ground-based engines, the company aims to double global travel speeds while eliminating the traditional sonic boom barrier.

How This 25-Year-Old Built A $675M Legal AI Startup (With No Legal Experience)

Gustaf Alströmer, Max Junestrand, Mark Mandelmann, Martin Splitt

Founded in 2024 by Max Giunestrand, Legora is an AI-powered workspace for legal professionals that recently secured an $80 million Series B funding round led by Iconic and General Catalyst. The company differentiates itself through a "software + service" hybrid model and a data-centric architecture that ensures GDPR compliance by hosting all European client data within the EU. With a flat organizational structure that scaled from ten to 100 employees in less than a year, Legora is aggressively expanding its operations to hubs in New York, London, and Stockholm to support law firms in navigating the AI transition.

Alexandr Wang: Building Scale AI, Transforming Work With Agents & Competing With China

Alexandr Wang, Jared

Meta has committed to investing over $14 billion in Scale AI, valuing the company at $29 billion while appointing Alexander Wang to lead its new AI superintelligence lab. Under Wang's direction, Scale is pivoting from its origins in data labeling to building agentic workflows for enterprise and government sectors, leveraging proprietary data and specialized fine-tuned models to differentiate in a fragmented market. The partnership aims to accelerate scientific breakthroughs and develop military applications like the Thunderforge system, positioning the company to manage a future economy where human operators oversee swarms of autonomous AI agents.

Inside The Hard Tech Startups Turning Sci-Fi Into Reality

Y Combinator advises hard tech founders to leverage a $500,000 investment and three-month timeline to validate specific technical kernels rather than building full-scale complex products. By isolating manageable milestones—such as securing massive Letters of Intent or demonstrating proof-of-concept prototypes—companies like Boom Supersonic and Relativity Space de-risk their ventures to achieve billion-dollar valuations or successful commercial launches. This strategy prioritizes "timeline compression" and tangible execution over traditional fundraising, allowing teams to mitigate technical risks while capitalizing on obvious market demand.

YC SUS: Eric Migicovsky & Dalton Caldwell discuss pivoting & pitching

Eric Migicovsky, Dalton Caldwell, Eric Majerjkowski, Rick Viscomi

This discussion outlines critical strategies for startup founders, emphasizing that pivots should be driven by a complete lack of traction and validated through concrete customer payments rather than theoretical debates. It further details how to effectively apply to Y Combinator by prioritizing clarity, avoid common pitfalls like over-relying on patents, and ensure sustainable unit economics by focusing on retention and revenue. Ultimately, the guidance stresses that success relies on building genuine product-market fit and maintaining financial flexibility to adapt quickly to external shocks.