SF Tensor: Training Infrastructure

• (01:23) - Will AWS Buy TPUs From Google?
• (20:51) - 𝕏 Timeline Reactions
• (46:33) - Harj Taggar, a Managing Partner at Y Combinator and co-founder of Triplebyte and Auctomatic, discusses the evolving landscape for startups, highlighting the increased ease of selling to both government entities and Fortune 500 companies. He emphasizes that the choice between these paths depends on the product type, noting that AI advancements have opened new opportunities for startups to secure large clients directly. Taggar also observes a trend where companies are adopting AI-native, full-stack approaches, integrating AI into their core operations rather than merely offering AI tools to existing firms.
• (59:39) - Richard Wang is co-founder and CEO of Clad Labs, a startup building “CHAD: The Brainrot IDE,” an AI-powered development environment designed to blend coding with leisure workflows.
• (01:06:32) - Philip Ho, Absurd is a San-Francisco–based startup that builds AI-powered brand and performance ads at scale.
• (00:00) - produce production-quality marketing videos scripted, generated, and edited by a multi-agent AI system in about 72 hours. Their work has already seen traction: one of their launch videos reportedly hit over 1 million views, and they average hundreds of thousands of organic views across their campaigns.
• (01:18:23) - Ali Attar, co-founder of Lightberry, discusses the company's mission to develop an operating system that enables humanoid robots to interact with humans through natural language, eliminating the need for coding. He highlights their collaboration with manufacturers like Unitree to integrate this software, allowing robots to perform tasks such as emceeing events autonomously. Attar also emphasizes the potential for diverse robot applications, including security roles, and envisions a future where robots are prevalent in public spaces, interacting seamlessly with people.
• (01:29:59) - Kurush Dubash, co-founder and CEO of Dome, discusses how Dome provides a unified API for prediction markets, enabling users and developers to trade and analyze data across multiple platforms simultaneously. He highlights that their clientele includes application developers, sports books, and hedge funds interested in high-frequency trading and internal pricing. Kurush also notes the increasing number of platforms entering the prediction market space, each targeting specific regions or verticals, and emphasizes Dome's role in aggregating fragmented liquidity to support professional traders.
• (01:38:40) - David Alade, co-founder of Sorce, introduces the app as a "Tinder for Jobs," where users upload their resumes, swipe right on job listings, and AI agents automatically complete applications on company websites. He discusses the current hiring market, noting that while inbound applications are still used, the process is ripe for disruption due to its inefficiencies. Alade also highlights Sorce's growth, mentioning over 25,000 interviews facilitated in the past year, and shares that their marketing strategy relies heavily on viral social media content, particularly on TikTok and Instagram.
• (01:45:23) - Karim Rahme, co-founder and CEO of Metorial, discusses how their platform enables AI agents to securely access various applications and data sources, such as Gmail, SAP, and Salesforce, while providing essential access control for large organizations. He highlights Metorial's open-source success, noting over 3,600 GitHub stars and nearly 1,000 weekly active users within five weeks of launch, and mentions ongoing discussions with Fortune 500 companies for large-scale deployment. Rahme also shares his background, including graduating from NYU Abu Dhabi in May and previously leading an Abu Dhabi-based ticketing startup for over three years.
• (01:52:54) - Michael Sakowski, co-founder and COO of Crunched, an AI software company, discusses how their Excel-native AI analyst is tailored for top finance professionals, distinguishing itself from Microsoft's broader Copilot by focusing on the specific needs of the top 1% of Excel users. He highlights Crunched's unique ability to detect errors in complex financial models, sharing an instance where the software identified a £10 million overvaluation in a private equity deal, thereby preventing significant financial misrepresentation. Sakowski also addresses concerns about data security, emphasizing that Crunched does not train on client data and cannot access user prompts, ensuring confidentiality for their clients.
• (02:01:00) - Nimit Maru, co-founder and CEO of Sava, discusses building an AI-powered trust company to modernize trust administration by treating the trust charter as programmable infrastructure, enabling efficient, compliant, and scalable services. He shares his experience with the outdated trust industry after selling his previous company, Fullstack Academy, and highlights how Sava's platform allows for real-time tracking and management of trusts, aiming to make sophisticated wealth planning more accessible.
• (02:08:34) - Ben Koska, co-founder of SF Tensor, discusses how their platform collaborates with various cloud providers to streamline AI model training by managing GPU allocations and optimizing for different hardware, allowing researchers to focus solely on their work. He emphasizes SF Tensor's exclusive focus on the training phase, addressing a gap in the market, and highlights the diverse clientele ranging from individual researchers to large-scale labs tackling unsolved problems in areas like drug discovery and protein folding. Koska also notes the potential for companies to enhance base models with proprietary data, indicating SF Tensor's capability to support such training needs.
• (02:15:51) - Henry Kwan, founder and CEO of Icarus, is an aerospace engineer with experience at NASA and Orbital, where he built drones and satellites. He discusses Icarus's development of solar-powered autonomous drones capable of flying at 60,000 feet for extended periods, offering advantages over satellites due to their proximity and cost-effectiveness. Initially targeting defense applications, Kwan envisions broader uses for these stratospheric drones, including enhanced connectivity and surveillance capabilities.
• (02:24:20) - Cole Dermott, co-founder of Locus, a Y Combinator-backed startup, discusses the company's development of payment infrastructure for AI agents, enabling them to autonomously pay for services while maintaining control through defined budgets and permissions. He highlights that initial adopters are developers creating autonomous agents capable of discovering and paying for services independently, with broader consumer adoption expected as trust in the technology grows. Dermott also shares that Locus has processed approximately 3,500 transactions and has around 80 projects built using their platform.
• (02:28:06) - Paul Graham & Jessica Livingston. Graham is an English-American computer scientist and entrepreneur, co-founded Y Combinator, a prominent startup accelerator that has funded over 3,000 startups, including Airbnb, Dropbox, Stripe, and Reddit. Jessica Livingston is a co-founder of Y Combinator and one of the most influential figures in modern startup culture. She helped build YC from a small experiment into the world’s most successful startup accelerator, backing companies like Airbnb, Stripe, Reddit, and Dropbox. Jessica is also the author of Founders at Work and a long-time advocate for early-stage founders.

TBPN.com is made possible by: 

Ramp - https://ramp.com

Figma - https://figma.com

Vanta - https://vanta.com

Linear - https://linear.app

Eight Sleep - https://eightsleep.com/tbpn

Wander - https://wander.com/tbpn

Public - https:...