Francesc Campoy
Showing 1–6 of 6 transcripts.
- Y Combinator50 min
Inside Claude Code With Its Creator Boris Cherny
Boris Cherny, Mark Mandelmann, Jr., Francesc Campoy, Melanie Warrick
Anthropic has accelerated its engineering velocity to a 150% growth rate by leveraging *Claude Code* to generate 70% to 90% of its own source code and rewriting its entire codebase every six months to avoid technical debt. The organization operates a bimodal workforce of specialists and generalists while predicting that traditional software engineering roles will fade as AI automates coding tasks. With *Claude Code* now powering the majority of developer sessions and influencing 4% of public industry commits, the company is simultaneously expanding into non-technical domains through *Co-work* and monitoring its safety rating at ASL 3.
- Y Combinator36 min
Anthropic Co-founder: Building Claude Code, Lessons From GPT-3 & LLM System Design
Tom Brown, Melanie Warrick, Mark Mandelbaum, Mark Mirchandani, Brian Dorsey, Priyanka Vergadia, Priyan Kavanagh, Leslie Kendrick, Francesc Campoy
Vivek Mirchandani, a founding engineer at OpenAI and Anthropic, pioneered the scaling laws strategy that established the link between compute investment and model intelligence before co-founding Anthropic to prioritize AI safety and alignment. Under his leadership, Anthropic shifted focus from early consumer prototypes to building robust training infrastructure, ultimately delivering the coding-dominant Claude 3.5 Sonnet model while navigating global bottlenecks in electricity and multi-vendor hardware supply chains. The organization's approach combines internal qualitative benchmarking with a multi-vendor hardware strategy to drive the largest infrastructure buildout in history, positioning the company at the center of the coming wave of AGI development.
- Jane Street1h 6m
The Uncertain Art of Accelerating ML Models with Sylvain Gugger
Sylvain Gugger, Ron Minsky, Jeremy Howard, Mark Mandelmann, Mark Mirchandani, Francesc Campoy, Gabriel Sanchez
Former fast.ai co-author Jeremy Howard discusses his transition from mathematics education to optimizing machine learning infrastructure at Jane Street, highlighting breakthroughs in learning rate schedules and image resizing that previously secured top benchmark placements. He details the development of the Hugging Face Accelerate library, a lightweight tool designed to abstract complex hardware parallelism and eliminate boilerplate code for training across diverse GPUs and TPUs. The discussion further explores the architectural constraints of financial data, the dominance of PyTorch's iterative execution model, and Jane Street's rigorous approach to reproducibility and custom model development for high-frequency trading.
- Y Combinator42 min
Lightcone: Consumer is back, What’s getting funded now, The vibes immaculate
Gary, Harge, Diana, Mark Mandelbaum, Mark Mandelmann, Mark Mirchandani, Mark Mandalini, Francesc Campoy, Jared Yanoski, Melanie Warrick, Dana
The Winter 2024 YC batch marks a historic platform shift where AI dominates 70% of 243 companies, driving total Annual Recurring Revenue from $6 million to $20 million while attracting a record number of MIT graduates. This cohort exhibits a distinct pivot toward consumer startups and developer infrastructure, reversing previous B2B and international expansion trends as founders prioritize tangible AI products over crypto or marketplaces. With median founder age dropping to 26 and 30% of startups pivoting to new ideas, the program positions itself at the foundational stage of an AI revolution comparable to 2007, signaling a massive opportunity to disrupt global software spending.
- Y Combinator1h 1m
Advice on Organizing and Running Growth Teams from Dan Hockenmaier and Gustaf Alströmer
Dan Hockenmaier, Gustaf Alströmer, Craig Cannon, Toni, Michael Savage, Justin LaRosa, Mark Mandelbaum, Francesc Campoy, Dan Galpin, Dan Gallup
Dan Hockenmeyer of Basis One and Gustav Alströmer of Y Combinator discuss a growth strategy centered on prioritizing a single high-leverage channel and achieving product-market fit before scaling paid acquisition. They argue that early-stage teams must rely on analytical generalists who can model the entire business equation, using retention graphs to test high-impact hypotheses rather than chasing superficial metric tweaks. The discussion highlights critical lessons from Airbnb and Thumbtack, emphasizing that sustainable expansion requires building growth functions in-house, treating referrals as paid marketing with strict ROI calculations, and avoiding structural distractions like premature VP hires or reliance on saturated ad platforms.
- Y Combinator2h 14m
IPFS, CoinList, and the Filecoin ICO with Juan Benet and Dalton Caldwell
Juan Benet, Dalton Caldwell, Craig Cannon, Dan Galpin, Francesc Campoy, Mark Mandelmann, Mark Blythouse, Mark Mandelbaum, Chris Broadfootnich
Protocol Labs, founded by Juan Bonet, is developing the InterPlanetary File System and its incentivized Layer-2, Filecoin, to transition internet infrastructure from centralized, location-based models to a decentralized, content-addressed network that ensures data sovereignty and censorship resistance. By replacing traditional venture capital with token-based funding models like CoinList, the organization enables long-term open-source research while utilizing Proof of Replication to create a liquid global market for unused storage capacity. This initiative aims to solve systemic inefficiencies in credit assignment and data redundancy, positioning Filecoin as a competitor to major cloud providers while fostering a new economy where protocol contributors are compensated directly through native network value.