Latent Space
Technology
About
swyx and Alessio on the AI engineer stack.
Episodes
- 🔬Doing Vibe Physics — Alex Lupsasca, OpenAI
Alex Lupsasca, from OpenAI, discusses the significant advancements in AI, specifically GPT-5, and its implications for theoretical physics. He shares his experiences of how AI has assisted in his research, including solving complex calcula…
- Physical AI that Moves the World — Qasar Younis & Peter Ludwig, Applied Intuition
Qasar Younis and Peter Ludwig of Applied Intuition discuss the development of physical AI, contrasting it with screen-based AI, and the advancement of autonomy tooling. They explore the company's mission to build physical AI for various mo…
- AIE Europe Debrief + Agent Labs Thesis: Unsupervised Learning x Latent Space Crossover Special (2026)
This Latent Space episode, a year after the first Unsupervised Learning x Latent Space Crossover special, discusses changes in the world of AI, recorded after AIE Europe and before the Cursor-xAI deal. Topics include AI infrastructure, the…
- Shopify’s AI Phase Transition: 2026 Usage Explosion, Unlimited Opus-4.6 Token Budget, Tangle, Tangent, SimGym — with Mikhail Parakhin, Shopify CTO
Mikhail Parakhin, Shopify CTO, explains how the company is integrating AI, discussing internal AI tool adoption, ML experimentation, and customer simulation. He also describes Shopify’s AI initiatives including Tangle, Tangent, and SimGym.
- 🔬 Training Transformers to solve 95% failure rate of Cancer Trials — Ron Alfa & Daniel Bear, Noetik
This episode of Latent Space features Ron Alfa and Daniel Bear from Noetik. They discuss how artificial intelligence, particularly transformer models like TARIO-2, can be applied to improve the success rates of cancer trials by optimizing…
- Notion’s Token Town: 5 Rebuilds, 100+ Tools, MCP vs CLIs and the Software Factory Future — Simon Last & Sarah Sachs of Notion
Simon Last and Sarah Sachs of Notion discuss the multi-year development and integration of Custom Agents into Notion. They delve into the repeated rebuilds, the shift to an agent-native system for enterprise work, and the product, engineer…
- Extreme Harness Engineering for Token Billionaires: 1M LOC, 1B toks/day, 0% human code, 0% human review — Ryan Lopopolo, OpenAI Frontier & Symphony
Ryan Lopopolo of OpenAI discusses harness engineering and the use of AI agents in software development. He explains how OpenAI’s Frontier team uses Codex to build and ship products with no human-written or human-reviewed code, highlighting…
- Marc Andreessen introspects on The Death of the Browser, Pi + OpenClaw, and Why "This Time Is Different"
Marc Andreessen joins Latent Space to discuss the evolution of AI, arguing that the current advancements signify a unique historical moment in technology. He shares his perspectives on why this AI cycle is distinct from previous booms and…
- Moonlake: Causal World Models should be Multimodal, Interactive, and Efficient — with Chris Manning and Fan-yun Sun
Chris Manning and Fan-yun Sun from Moonlake AI discuss their approach to Causal World Models. They focus on multimodal, interactive, and efficient models, contrasting them with other existing world models like Genie 3.
- Mistral: Voxtral TTS, Forge, Leanstral, & what's next for Mistral 4 — w/ Pavan Kumar Reddy & Guillaume Lample
In this episode, Pavan Kumar Reddy and Guillaume Lample of Mistral discuss Voxtral TTS, Forge, Leanstral, and what is next for Mistral 4. They also delve into the architecture and open research behind Mistral's models.
- 🔬Why There Is No "AlphaFold for Materials" — AI for Materials Discovery with Heather Kulik
This episode of Latent Space features Professor Heather Kulik discussing the application of AI in materials discovery. She emphasizes the importance of integrating domain expertise with AI techniques to achieve successful outcomes in the f…
- Dreamer: the Personal Agent OS — David Singleton
David Singleton, formerly of Stripe, discusses Dreamer, a new consumer-first platform for AI agents and agentic applications. Dreamer allows users to discover, build, and use AI agents, centered on a personal "Sidekick".
- Why Anthropic Thinks AI Should Have Its Own Computer — Felix Rieseberg of Claude Cowork & Claude Code Desktop
Felix Rieseberg discusses why Anthropic believes AI should have its own computer, focusing on the development of Claude Cowork for knowledge work. He also shares insights into building impactful desktop apps and the evolving landscape of A…
- Retrieval After RAG: Hybrid Search, Agents, and Database Design — Simon Hørup Eskildsen of Turbopuffer
Simon Hørup Eskildsen discusses the development of Turbopuffer, a search engine for unstructured data. He covers the company's origins in addressing the high costs of vector search for Readwise and its architectural choices, including the…
- NVIDIA's AI Engineers: Agent Inference at Planetary Scale and "Speed of Light" — Nader Khalil (Brev), Kyle Kranen (Dynamo)
Nader Khalil and Kyle Kranen discuss NVIDIA's AI developments. The conversation covers agent inference at scale, the Dynamo data center inference engine, and the Brev platform for GPU access. They also touch upon Jensen's "Speed of Light"…
- Cursor's Third Era: Cloud Agents
Latent Space discusses Cursor's cloud agents, specifically focusing on cloud agents for computer use, slash commands, and subagents. The episode also covers personalized coding models and the future of continual learning.
- Every Agent Needs a Box — Aaron Levie, Box
Latent Space features Aaron Levie, CEO of Box, to discuss why every AI agent needs a box. They explore topics such as agent governance, identity, coding agents, and read/write agent workflows.
- METR’s Joel Becker on exponential Time Horizon Evals, Threat Models, and the Limits of AI Productivity
Joel Becker from METR is featured on the Latent Space podcast, discussing exponential time horizon evaluations, threat models, and the limits of AI productivity. The episode also delves into METR’s research on developer productivity.
- [LIVE] Anthropic Distillation & How Models Cheat (SWE-Bench Dead) | Nathan Lambert & Sebastian Raschka
This episode features Nathan Lambert and Sebastian Raschka discussing Anthropic distillation and how models might exhibit cheating behavior on SWE-Bench. It is a live recording shared with subscribers.
- 🔬Searching the Space of All Possible Materials — Prof. Max Welling, CuspAI
Professor Max Welling discusses CuspAI, a company focused on AI-driven materials discovery. He connects quantum gravity, equivariant neural networks, and diffusion models to climate-focused materials, describing a "physics processing unit"…
- Claude Code for Finance + The Global Memory Shortage: Doug O'Laughlin, SemiAnalysis
This episode features Doug O'Laughlin of SemiAnalysis, discussing the use of Claude Code for financial analysis. The conversation also covers the global memory shortage and its potential impact.
- ⚡️The End of SWE-Bench Verified — Mia Glaese & Olivia Watkins, OpenAI Frontier Evals & Human Data
Mia Glaese and Olivia Watkins discuss OpenAI's decision to stop evaluating SWE-Bench Verified due to saturation and contamination. They explain the shift to SWE-Bench Pro for coding benchmarks.
- Bitter Lessons in Venture vs Growth: Anthropic vs OpenAI, Noam Shazeer, World Labs, Thinking Machines, Cursor, ASIC Economics — Martin Casado & Sarah Wang of a16z
Martin Casado and Sarah Wang of a16z discuss how venture and growth capital are merging in the AI space. They cover the financing strategies for AI companies, including compute contracts and rapid capability growth, and examine the potenti…
- Owning the AI Pareto Frontier — Jeff Dean
Jeff Dean, Chief AI Scientist at Google, discusses his extensive work in AI, including his contributions to Google Search, Pareto frontier strategy, and distillation techniques. He also touches on topics such as sparse models, unified vs.…
- 🔬Beyond AlphaFold: How Boltz is Open-Sourcing the Future of Drug Discovery
This podcast discusses the evolution of structural biology models, including AlphaFold, Boltz-1, and Boltz-2. The co-founders of Boltz explain their mission to open-source and democratize drug discovery through these models.
- The First Mechanistic Interpretability Frontier Lab — Myra Deng & Mark Bissell of Goodfire AI
Mark Bissell and Myra Deng of Goodfire AI discuss their work on mechanistic interpretability, focusing on building APIs and managing enterprise deployments. They explain Goodfire’s method for enhancing the AI lifecycle by creating an inter…
- 🔬 Automating Science: World Models, Scientific Taste, Agent Loops — Andrew White
Andrew White discusses his experiences with AI in scientific discovery, focusing on topics such as ChemCrow, the concept of scientific taste, and autonomous research systems including Kosmos.
- Captaining IMO Gold, Deep Think, On-Policy RL, Feeling the AGI in Singapore — Yi Tay
Yi Tay discusses his work on Gemini Deep Think and IMO Gold at Google DeepMind, focusing on the company's shift towards RL-driven reasoning and the development of large-scale AI models.
- Brex’s AI Hail Mary — With CTO James Reggio
James Reggio, CTO of Brex, discusses the company's AI strategy, including corporate, operational, and product AI. He explains how Brex approaches AI transformation within a regulated financial institution, focusing on compliance, auditabil…
- Artificial Analysis: Independent LLM Evals as a Service — with George Cameron and Micah-Hill Smith
George Cameron and Micah-Hill Smith of Artificial Analysis discuss their independent LLM evaluation service. They explain how they conduct comprehensive evaluations across various models and provide insights to developers and enterprises.
- [State of Evals] LMArena's $1.7B Vision — Anastasios Angelopoulos, LMArena
Anastasios Angelopoulos of LMArena discusses the company's recent $150M Series A funding, valuing them at $1.7B. He shares insights into LMArena's growth as an AI evaluation platform, its operational strategies, and its future vision for a…
- [NeurIPS Best Paper] 1000 Layer Networks for Self-Supervised RL — Kevin Wang et al, Princeton
Kevin Wang and his team from Princeton discuss their NeurIPS Best Paper on scaling reinforcement learning networks to 1,000 layers deep. They explain how self-supervised RL and architectural innovations like residual connections and layer…
- [State of Code Evals] After SWE-bench, Code Clash & SOTA Coding Benchmarks recap — John Yang
John Yang, creator of SWE-bench, discusses its evolution into a standard for evaluating AI coding agents. He also covers CodeClash and other benchmarks, touching on topics like verification methods, data challenges, and the potential for h…
- [State of Post-Training] From GPT-4.1 to 5.1: RLVR, Agent & Token Efficiency — Josh McGrath, OpenAI
Josh McGrath from OpenAI discusses the evolution of post-training, including RLVR, agent efficiency, and token efficiency in models such as GPT-4.1 and GPT-5.1. He shares insights on his work from pre-training data curation to shipping var…
- [State of RL/Reasoning] IMO/IOI Gold, OpenAI o3/GPT-5, and Cursor Composer — Ashvin Nair, Cursor
Ashvin Nair, formerly of OpenAI and now leading model development at Cursor, shares insights into the evolution of AI reasoning and the challenges of RL. He details the growth of OpenAI's reasoning team and Cursor's unique approach to cont…
- [State of AI Startups] Memory/Learning, RL Envs & DBT-Fivetran — Sarah Catanzaro, Amplify
Sarah Catanzaro, from Amplify Partners, discusses the current landscape of AI startups. She covers topics such as funding trends, the use of data infrastructure by frontier AI labs, and the future of AI applications.
- One Year of MCP — with David Soria Parra and AAIF leads from OpenAI, Goose, Linux Foundation
This episode of Latent Space explores the first year of the Model Context Protocol (MCP), an open standard launched by Anthropic to connect AI applications. It details MCP's growth, adoption by major tech companies, and its transition to t…
- Steve Yegge's Vibe Coding Manifesto: Why Claude Code Isn't It & What Comes After the IDE
Steve Yegge talks about 'vibe coding' and the future of software development beyond Integrated Development Environments (IDEs). He discusses agent orchestration, managing AI agents, and his perspectives on the evolving landscape of coding…
- ⚡️GPT5-Codex-Max: Training Agents with Personality, Tools & Trust — Brian Fioca + Bill Chen, OpenAI
Bryan Fioca and Bill Chen from OpenAI discuss GPT5-Codex-Max, a new long-running coding agent. They delve into how it is trained with personality, tools, and trust, and its ability to architect, refactor, and ship entire features. The disc…
- SAM 3: The Eyes for AI — Nikhila & Pengchuan (Meta Superintelligence), ft. Joseph Nelson (Roboflow)
This episode of Latent Space features Nikhila Ravi and Pengchuan Zhang discussing SAM 3. They explain how SAM 3 unifies interactive segmentation, open-vocabulary detection, and video tracking into a single model, and its applications in va…
- ⚡️Jailbreaking AGI: Pliny the Liberator & John V on Red Teaming, BT6, and the Future of AI Security
Pliny the Liberator and John V join Latent Space to discuss their work in AI red-teaming and open-source AI security. They delve into jailbreaking AGI models, strategies like universal jailbreaks and multi-turn crescendo attacks, and their…
- AI to AE's: Grit, Glean, and Kleiner Perkins' next Enterprise AI hit — Joubin Mirzadegan, Roadrunner
This episode features Joubin Mirzadegan, who shares his experiences in go-to-market strategies and building companies. He discusses topics such as the origins of the CRO-only podcast Grit, his work with Glean and Windsurf, and his new comp…
- The Future of Email: Superhuman CTO on Your Inbox As the Real AI Agent (Not ChatGPT) — Loïc Houssier
LoĂŻc Houssier, CTO of Superhuman, discusses his journey from cryptography to AI in email. He details how Superhuman integrates AI without adding latency, utilizes agentic search, and envisions the inbox as a future AI agent.
- World Models & General Intuition: Khosla's largest bet since LLMs & OpenAI
This episode discusses world models and General Intuition, a company founded by Pim that trains AI agents using human gameplay data. It explores how these models, which interpret visual data and output actions, can transfer from games to r…
- After LLMs: Spatial Intelligence and World Models — Fei-Fei Li & Justin Johnson, World Labs
Fei-Fei Li and Justin Johnson, co-founders of World Labs, discuss their new generative "world model" called Marble, which creates editable 3D environments from text and images. They explain how spatial intelligence is the next frontier aft…
- ⚡️ 10x AI Engineers with $1m Salaries — Alex Lieberman & Arman Hezarkhani, Tenex
Alex Lieberman and Arman Hezarkhani, co-founders of Tenex, explain their model of compensating AI engineers based on output, rather than hours. This approach has led to significant productivity gains and high compensation for engineers.
- Anthropic, Glean & OpenRouter: How AI Moats Are Built with Deedy Das of Menlo Ventures
Deedy Das of Menlo Ventures discusses the growth of Anthropic and Glean, and how AI is impacting enterprise software and coding. He also shares insights on AI infrastructure and research investing, and the future of engineering.
- ⚡ Inside GitHub’s AI Revolution: Jared Palmer Reveals Agent HQ & The Future of Coding Agents
Jared Palmer, SVP at GitHub and VP of CoreAI at Microsoft, discusses the evolution of coding agents and developer tools. He provides insights into GitHub Universe and the launch of Agent HQ, a new collaboration hub for coding agents and de…
- ⚡ [AIE CODE Preview] Inside Google Labs: Building The Gemini Coding Agent — Jed Borovik, Jules
Jed Borovik, Product Lead at Google Labs, joins Latent Space to discuss Google's AI-powered software development. He shares insights into the Gemini coding agent, Jules, and its evolution from internal tools to public products.
- ⚡️ Ship AI recap: Agents, Workflows, and Python — w/ Vercel CTO Malte Ubl
Vercel CTO Malte Ubl discusses the company's AI-powered development infrastructure, including agents, workflows, and Python support. He shares insights into Vercel’s approach to building and deploying AI tools and their vision for the futu…