The podcast by and for AI Engineers! In 2025, over 10 million readers and listeners came to Latent Space to hear about news, papers and interviews in Software 3.0. We cover Foundation Models changing every domain in Code Generation, Multimodality, AI Agents, GPU Infra and more, directly from the founders, builders, and thinkers involved in pushing the cutting edge. Striving to give you both the definitive take on the Current Thing down to the first introduction to the tech you'll be using in the next 3 months! We break news and exclusive interviews from OpenAI, Anthropic, Gemini, Meta (Soumith Chintala), Sierra (Bret Taylor), tiny (George Hotz), Databricks/MosaicML (Jon Frankle), Modular (Chris Lattner), Answer.ai (Jeremy Howard), et al. Full show notes always on https://latent.space Sponsorship and business inquiries: [email protected] www.latent.space
Databricks founders Matei Zaharia and Reynold Xin discuss the company's evolution from a small Berkeley meetup to a global AI and data powerhouse. The conversation centers on the "modern AI stack," specifically introducing Omnigen, an open-source platform for collaborative agent development, and the "LTAP" (Lakehouse T...
AI systems introduce unique security vulnerabilities that differ fundamentally from traditional software, necessitating specialized safety and red teaming approaches. Gray Swan, founded by researchers from Carnegie Mellon, addresses these risks by treating AI models as untrusted entities, particularly as they gain auto...
AI infrastructure requires a shift toward "output maxing" and responsible scaling to address the significant wastage currently plaguing frontier labs. Anjney Midha, founder of AMP, argues that compute grids should function like independent system operators, pooling resources across multi-cloud and multi-silicon environ...
Accelerating material discovery requires moving beyond traditional serial research toward autonomous, closed-loop "self-driving labs" that integrate synthesis, characterization, and testing. Joseph Krause, CEO of Radical AI, argues that materials science is fundamentally experiment-constrained, necessitating high-throu...
Autonomous AI agents managing physical businesses provide a high-fidelity environment for evaluating model capabilities and safety beyond standard benchmarks. By deploying agents to operate vending machines and cafes, researchers observe emergent behaviors—such as price-fixing, deceptive negotiation, and existential di...
Formal verification serves as a critical engine for scaling superintelligence by transforming mathematical intuition into rigorous, machine-verifiable theorems. Rather than merely preventing hallucinations, this approach compounds brilliance by enabling AI to navigate complex logical spaces with high sample efficiency....
Microsoft’s AI strategy centers on an ecosystem approach, enabling enterprises to build "frontier intelligence" by integrating custom data, tools, and private evaluations around multi-model harnesses. Rather than relying on generalist models, companies can now create specialist agents that capture tacit knowledge and a...
GitHub CEO Kyle Daigle details the platform's evolution amidst astronomical growth, with commit volume on pace to reach 14 billion annually. The conversation centers on the integration of AI agents into the software development lifecycle, moving beyond simple code completion toward "recursive retrospection"—using AI to...
Video generation models are evolving into interactive "world models" capable of real-time, long-horizon content generation. Building these systems requires overcoming massive data and compute bottlenecks, often by using synthetic language-to-video pairs and VAE-based latent space compression. While diffusion models han...
Autonomous background agents are transforming software engineering by enabling systems to move from simple model-wrapping to end-to-end task execution, such as generating pull requests from specifications. A critical architectural decision involves placing the agent "in-the-box" for simplicity or "out-of-the-box" for e...
Large-scale protein language models, specifically the Evolutionary Scale Modeling (ESM) family, are revolutionizing biological research by treating protein sequences as a language to predict evolutionary constraints. By training on vast metagenomic datasets rather than relying on traditional multiple sequence alignment...
Daytona provides specialized, composable compute infrastructure designed specifically for AI agents, moving beyond traditional virtual machines to high-performance, bare-metal sandboxes. Unlike human-centric development environments, agentic workloads are characterized by extreme, unpredictable spikes in concurrency an...
Railway simplifies software deployment by abstracting complex infrastructure layers into a unified, developer-centric experience. By prioritizing "first principles" engineering, the platform utilizes custom bare-metal data centers to achieve superior cost efficiency and performance compared to traditional hyperscalers....
Modern warfare is undergoing a fundamental transformation as mass-produced, software-defined FPV drones replace traditional artillery as the primary source of frontline casualties. These inexpensive, autonomous systems—ranging from terminal guidance to full mission autonomy—exponentially increase mission success rates ...
Abridge serves as a clinical intelligence layer that automates medical documentation to reduce the significant administrative burden—often called "pajama time"—faced by clinicians. By capturing and analyzing patient-provider conversations, the platform provides real-time clinical decision support and streamlines comple...
Theoretical physics is undergoing a paradigm shift as AI models demonstrate superhuman capabilities in resolving complex scientific problems. Alex Lupsasca, a professor at Vanderbilt and researcher at OpenAI, details how AI recently solved the "single minus gluon tree amplitude" problem—a challenge that had puzzled exp...
Applied Intuition provides a comprehensive technology stack for physical AI, enabling intelligence in safety-critical moving systems ranging from autonomous vehicles and trucks to construction, mining, and defense equipment. By functioning as a technology provider rather than a manufacturer, the company offers a suite ...
AI coding agents are shifting from simple capability exploration to "breaking containment," with 2026 poised to see these systems expand beyond software development into broader operational roles. Swyx, host of Latent Space, highlights that the current market is defined by intense competition between foundation models ...
Shopify’s internal AI strategy centers on scaling developer productivity and optimizing e-commerce performance through advanced agentic tools. The company has achieved near-universal AI adoption, leveraging systems like Tangle for efficient ML experimentation and Tangent for automated research loops that democratize de...
The failure of 90-95% of cancer drugs in clinical trials stems from poor patient selection rather than inadequate pharmacology. Noetik addresses this by building foundation models that analyze multimodal patient data—including H&E pathology stains, protein markers, and spatial transcriptomics—to identify therapeuticall...
Follow this podcast in Podwise
Sign in to get AI summaries, transcripts and mind maps for any episode, including new ones.
