Latent Space

Swyx joins SAIL Live with Nathan Lambert and Sebastian Raschka to discuss the current state of AI reasoning, comparing approaches from OpenAI, Anthropic, and open-source alternatives.

Felix Muller and the Anthropic team discuss Claude Code and Claude Cowork, tools that represent a shift from AI-assisted coding to AI-native development workflows.

Swyx's comprehensive review of 2025 in AI: the rise of reasoning models, the agent explosion, the open-source surge, and the economic realities catching up to the hype.

Ryan Lopopolo from OpenAI's Frontier team discusses how they run a 1M+ line-of-code codebase with zero human-written code, pioneering harness engineering as a discipline.

Deep dive into the economics of AI infrastructure: compute costs, inference pricing trends, and when it makes sense to build versus buy AI capabilities.

The creators of Anthropic's Model Context Protocol discuss the open standard for connecting AI assistants to external tools, now adopted by OpenAI and Google.

Swyx recaps the AI Engineer Summit 2025, identifying agents as the dominant theme and mapping the emerging stack for multi-agent systems.

Marc Andreessen argues AI is not a hype cycle but the culmination of 80 years of computer science research. The payoff was always coming — the question was when, not if.

Chris Lattner discusses how Modular is matching NVIDIA's CUDA performance on AMD hardware, potentially breaking the GPU monopoly that constrains AI development.

Alessio synthesizes lessons from investing in dozens of AI agent startups, identifying which patterns work in production and which remain research curiosities.

Examining Anthropic's approach to AI safety from a technical perspective: Constitutional AI, RLHF, and the emerging safety stack that shapes how Claude behaves.

Jeff Dean discusses how Google DeepMind is rethinking the AI infrastructure stack from hardware through frameworks to applications, and what the next generation of AI systems will look like.

David Hershey from Anthropic discusses how Claude Plays Pokemon began as a tool for experimenting with AI agents and revealed unexpected insights about long-horizon agent behavior.

Tracking the evolution of software engineering agents from Devin through the current generation, evaluating which approaches are working and where the field is heading.

Max Welling discusses CuspAI and how AI is accelerating materials science discovery, potentially unlocking new battery chemistries, superconductors, and carbon capture materials.

How Cursor and similar AI-native IDEs are reshaping the developer experience by making AI a first-class part of the coding environment rather than a bolted-on feature.

Examining how small, efficient models are finding product-market fit in edge computing, mobile devices, and cost-sensitive applications where frontier models are impractical.

Mapping the emerging AI security landscape: prompt injection, data poisoning, model theft, and the new attack surfaces created by AI agents that act autonomously.

A comprehensive overview of the open-source AI landscape in 2025, mapping which models, tools, and frameworks are gaining traction and which are fading.

Swyx examines the vibe coding phenomenon — building software by describing what you want in natural language — and whether it represents progress or a dangerous abstraction.

A practitioner's guide to Retrieval Augmented Generation in 2025: which patterns work in production, which are overhyped, and where the technology is headed.

OpenAI co-founder Greg Brockman discusses the dramatic boardroom coup that briefly ousted Sam Altman in November 2023. He provides an insider's perspective on the tensions between AI safety governance and rapid commercialization.

Andrej Karpathy provides a comprehensive overview of how LLMs work, what they can and cannot do, and where the field is heading. He frames LLMs as a new kind of operating system that will mediate between humans and all digital information.

Chris Lattner discusses why AI software infrastructure is fundamentally broken and how Modular and the Mojo programming language aim to fix it. He draws on his experience building LLVM and Swift to explain why the AI stack needs to be rebuilt from the ground up.

George Hotz discusses his vision for tinycorp: making AI compute accessible to everyone by building open-source hardware and software that commoditizes petaflop-scale computing. He argues that NVIDIA's monopoly on AI compute is the single biggest bottleneck in AI progress.

Simon Willison discusses the leaked Google memo 'We Have No Moat' and what open-source AI means for the future of the industry. He argues that open-source models are catching up to closed models faster than anyone expected.