All episodes

🔬 Why Transformers Hit a Wall the Moment Physics Shows Up — Anima Anandkumar, Caltech

Aug 26, 2026 · 1:23:32

Anima Anandkumar, Caltech professor and ex-NVIDIA lead, argues physical-world modeling needs neural operators that bake in physics, not language-style scaling; it already rivals supercomputers. Physics-informed nets fail; Fourier neural operators capture multi-scale non-local phenomena, and FourCastNet, trained on 50,000 reanalysis samples, forecasts weather tens of thousands of times faster on a consumer GPU. FourCastNet 3’s spherical harmonics keep month-long rollouts stable for ensemble climate prediction; the same operators build a fusion digital twin a million times faster than simulation and inverse design. TorchLean ports PyTorch to Lean for formal robustness proofs, and on the UN Science Advisory Board she urges AI for science not be regulated like chatbots. Bottleneck: more compute.

⏭️ Forward Deployed: Voice AI on what works in 2026

Aug 25, 2026 · 36:31

Basil Chatha hosts a fireside chat with voice-AI leaders Basia Sudol (Decagon), Varun Singh (Daily), Steven Diaz (Vapi), Tyler D'Silva (Retell AI), and Sudarshan Kamath (Smallest AI) arguing that production voice agents still run on cascaded STT→LLM→TTS pipelines, not end-to-end speech-to-speech. Varun explains why outbound calls are easier than inbound. Steven and Tyler defend cascaded for guardrails and context optimization; Sudarshan says Smallest's Hydra is multimodal but enterprises still prefer cascaded hybrids. On latency, Tyler says fillers make waits natural, Steven warns giant prompts hurt cost, and Varun suggests specialized sub-agents. They also cover multilingual TTS swapping, model choices, and open-source evals.

Simulating Humanity: from Generative Agents to 8 Billion Digital Twins — Joon Sung Park, Simile AI

Aug 21, 2026 · 1:11:01

Joon Sung Park, co-founder and CEO of Simile AI, joins Latent Space to argue that simulating humans requires bespoke behavior models, not just frontier LLMs, a bet rooted in his Generative Agents/Smallville paper. Frontier models optimized to be rational miss how people actually err, scoring 20–30% on niche populations, whereas Simile's digital twins match people's own responses 85% of the time. He details the three data buckets—interviews, observational data, and randomized controlled trials—and shows how companies use the platform for concept testing, A/B tests, and earnings calls. Park also shares the scaling-law glimpse in simulation, the economics of data-center-scale worlds, and his vision of simulating all 8 billion people to tackle wicked problems like climate change and UBI, with Thomas Schelling's agent-based models as precedent.

Exo: Harnesses should see their own code and logs — Alex Krentsel, UC Berekeley / Google Research

Aug 15, 2026 · 47:11

Alex Krentsel presents Exo, a fully recursive self-improving agent harness he built with Martín Casado and Anker Goyal, arguing agents can now safely edit their own code and runtime policy. Exo's architecture splits an agent into a stateless executor, a stateful harness with history and secrets, and an isolated sandbox, letting it change every component, roll back automatically, and teleport state—unlike OpenClaw/Pi, which only allow memory, skills, and tool extensions. The episode details Exo cutting its Discord adapter costs by 96% at runtime, plus secret stores and evals to prevent reward hacking. Krentsel argues RSI is newly possible because the harness is code in the same medium as LLM output; Exo runs in production at Brain Trust, with contributors directed to github.com/exoharness/exo.

🔬They Thought the Model Was Broken — Matt McPartlon & Neil Patil, Chai Discovery

Aug 11, 2026 · 1:35:20

Matt McPartland and Neil Patil of Chai Discovery explain why pharma partners like Eli Lilly, Pfizer, Novartis, and argenx are buying their AI design tools, arguing that Chai-2's all-atom diffusion model crossed a usefulness threshold by designing antibodies to 50 targets with hits on half at ~20% hit rate. They trace the lineage from open-sourced Chai-1 structure prediction through Chai-2's co-design of sequence and structure, and detail how validation via cryo-EM achieved a 0.33 Å error, making them suspect data leakage. The product resembles SolidWorks or Figma, not ChatGPT, with paint-tool epitope selection and content-aware fill for binder generation, built under strict pharma IP constraints via single-tenancy. They discuss convincing skeptical scientists—one cried after seeing a binder for a target she'd spent a decade on—and argue the compute market is mispriced for this model class, with bottlenecks in validation loops and talent obscurity rather than data or compute.

Next 100x in AI: Inference, Networking, & Self-Optimizing Models — Philip Kiely & Ali Taha, Baseten

Aug 3, 2026 · 1:42:54

Philip Kiely and Ali Taha of Baseten join Swyx to explain what actually happens when a 200,000-token request hits a production inference system, arguing that stacking quantization, speculative decoding, and disaggregated prefill/decode can make open models like GLM-5.2 up to 10x faster. They detail cache-aware routing, training traffic-specific speculators, and why quantization errors can cancel out so a more-quantized model beats a less-quantized one. They also reveal how Baseten grafted Kimi's vision encoder onto GLM-5.2, discuss NVIDIA Dynamo as a toolkit rather than a turnkey speedup, and explain why they are bearish on mega kernels. The conversation covers video generation's compute barriers, the trend toward ASIC-like GPUs, and the emerging loop where GLM-5.2 writes the GPU kernels that serve itself.

OpenAI’s Vision for the AI Super App — Akshay Nathan, OpenAI

Jul 28, 2026 · 1:10:57

Akshay Nathan, OpenAI's head of core product engineering, explains why the company merged Codex and ChatGPT into a unified agent harness, arguing that the lines between software engineering and knowledge work are blurring. He reveals that Codex unexpectedly gained traction among non-developers at OpenAI, leading to the launch of ChatGPT Work, which shares the same underlying harness but offers opinionated UX differences like sandboxing and Git visibility. Nathan details how artifacts, Sites, plugins, memory, and sub-agents enable users to replace slide decks and spreadsheets with interactive websites, and shares that ChatGPT Work reached 10 million users shortly after launch. He discusses OpenAI's goal to bring useful agents from developers to all knowledge workers, emphasizing that the default model configuration should work for most users while power users can opt into deeper reasoning, Ultra, or multi-agent setups. Nathan also reflects on how AI transforms product development, noting that ideas and taste become the bottleneck when anyone can build, and warns managers not to conflate motion with progress.

The AI Frontier: from open weights to open research — Eiso Kant, Poolside AI

Jul 22, 2026 · 1:56:13

Poolside co-founder Eiso Kant argues that open weights and open research are essential for a future with many foundation model companies, and that models like their new Laguna S (118B parameters, 8B active) can achieve remarkable persistence and capability through post-training reinforcement learning. He explains how Poolside's Model Factory enables 5-8 week model cycles with 10,000-20,000 experiments per month, streaming data directly into training, and perfect reproducibility. Kant shares why Poolside embraced open source after starting as a closed company, criticizes MCP and tool calls in favor of models writing code directly, and predicts reinforcement learning will move earlier into pre-training. He also calls on researchers to start new foundation model companies to avoid an oligopoly of intelligence.

🔬Causal Models Need Causal Data - Xaira’s X-Cell model (Bo Wang & Ci Chu)

Jul 21, 2026 · 1:29:47

Bo Wang and Ci Chu from Xaira Therapeutics present X-Cell, a 4.9-billion-parameter diffusion language model trained on the largest genome-wide CRISPRi Perturb-seq dataset (25.6 million single cells, 16 biological contexts) that predicts cellular responses to genetic perturbations and generalizes from immortalized cell lines to primary T cells from real donors. They explain why observational atlases describe biology but can't predict interventions, why they abandoned autoregression for a diffusion 'editing' approach, and how a model trained on immortalized cells predicted perturbation responses in primary T cells. They highlight a counterintuitive scaling result: X-Cell scales like an LLM on training loss, but generalization is bottlenecked by data diversity, not compute. The episode also covers Xaira's three-platform strategy (protein design, virtual cell, patient representation) and the central thesis that for causal models, the hard part isn't the model, it's the data.

🔬 RL with Verifiable Rewards, but the Verifier is a Lab — Lila Sciences

Jul 16, 2026 · 1:41:04

Andy Beam (CTO) and Rafa Gómez-Bombarelli (Co-founder & CSO of Physical Sciences) of Lila Sciences argue that science is an 'infinite token generator' for AI, using reinforcement learning with verifiable rewards where the wet lab acts as the verifier. They claim one general model trained on ~10 trillion experimentally-verified reasoning tokens across biology, chemistry, and materials outperforms domain-specific models—'breadth gives us depth.' Their AI Science Factories treat the lab as a data center, with instruments on a 'PCI bus' and humans 'below the API line.' Highlights include a CAR-T candidate designed in six months by two or three people, 'monster UTRs' achieving ~10x Moderna/Pfizer mRNA expression, and the 'zero-FTE startup' business model. They discuss RL pathologies like collapsed chains of thought and a model that 'swears,' and note why there is still no AlphaFold for materials due to the sim-to-real gap.

The AI Memory Problem: Why Long Context Isn’t Enough — Dan Biderman, Engram Co-founder & CEO

Jul 13, 2026 · 49:44

Engram co-founder and CEO Dan Biderman tells hosts Allen Park and Sean that long context and RAG aren't enough for AI memory—continual learning and gradient-based weight updates are needed. He argues Engram's approach compresses company knowledge into 'cartridges' that let models reason with far fewer tokens, overcoming 'context rot' and the inefficiency of re-reading massive corpora. Biderman draws on his background in Israeli special forces and computational neuroscience to explain how training creates intuition beyond text retrieval, citing examples like Harvey's legal queries where holistic understanding beats search. He envisions personal AI weights that improve like Tamagotchis, driven by user-specific feedback loops, and stresses that token efficiency and intelligence are inseparable—doing more with less enables harder problems. Engram is hiring infrastructure engineers to deploy millions of continuously updated memories.

Why AI Agents Don't Actually Understand You — Danielle Perszyk, Amazon AGI Lab

Jul 11, 2026 · 48:54

Danielle Perszyk of Amazon's AGI Lab argues that AI agents fail because they don't truly understand users — they model tasks but not the human mind, so reliability must shift from clicking correctly to aligning representations. She explains how Amazon's AGI Lab, seeded by the former Adept team, builds perception agents that perceive digital environments like humans and interact in real time, moving beyond chatbots and coding agents. The lab is exploring new architectures for episodic memory, social world models, and multi-agent systems where agents fluidly negotiate meaning, inspired by human collective intelligence. Perszyk warns that current AI homogenizes thinking, reducing human agency, and advocates for a diverse society of AIs with different biases to augment rather than replace human cognition. She discusses how aligning AI's goals with inferring and matching human representations could unlock generalization and make agents genuinely collaborative, from automating digital drudgery to transforming education with Socratic tutors.

Podcast Crossover: AIE, AGI, frontier lab strategy with ​ ⁨@matthew_berman⁩ and @swyxtv

Jul 10, 2026 · 28:03

Shawn 'Swyx' Wang, founder of the AI Engineer conference, tells Matthew Berman how he seized the industry shift by buying ai.engineer and partnering with a veteran conference organizer, crediting Andrej Karpathy's early endorsement. He argues Etched's ASICs are a natural next-gen bet for transformer inference, not an NVIDIA disruptor, and that Fable 5's slowness and cost signal the end of the LLM scaling era—making model efficiency the next problem. Swyx interprets OpenAI's reported 5% equity offer to the US government as a pragmatic multi-turn negotiation, likening it to Singapore's Temasek model, but warns against premature utility regulation. He pegs his own P(doom) at ~5% over 50 years, rejecting near-term doomerism as egotistical. For founders, he advocates building 'agent labs' that solve specific customer problems (e.g., for lawyers, dentists) rather than betting on model routing, which he dismisses as a marketing line that fails to exploit a single model's full stack as deeply as frontier labs do.

The Future of AI Infra: from Kubernetes to Agent Sandboxes — Akshat Bubna, Modal CTO

Jul 8, 2026 · 59:10

Modal CTO Akshat Bubna explains how the cloud platform evolved from a serverless runtime to an AI cloud built for elastic inference, agent sandboxes, and post-training workloads. He argues that Kubernetes was never designed for bursty GPU-heavy AI workloads, so Modal built a decorator-based infrastructure that co-locates compute requirements with code. The company added GPUs a year before ChatGPT and now powers inference for custom models at Suno, Runway, and robotics firms, using GPU snapshotting to slash cold starts. Modal's new Auto Endpoints incorporate open-source DeFlash speculative decoding for frontier-level performance. For reinforcement learning rollouts, Modal provides up to 100,000 sandboxes simultaneously. The platform spans 17 cloud providers, features private IPv6 networking via eBPF, and supports serverless multi-node training with RDMA. Bubna also reveals Modal's shift from developer experience to agent experience, noting that agents benefit from the same minimalist SDK and observability tools that humans use.

🔬 "The Most Innovative Diffusion Research Is Happening in Drug Discovery, Not Image Generation"

Jun 30, 2026 · 1:48:40

Evan Feinberg and Sergey Edunov of Genesis Molecular AI argue that diffusion models have unlocked sub-ångström accuracy in protein-ligand structure prediction, a breakthrough that makes AI useful for real drug discovery where the field’s favored 2Å RMSD benchmark is "slop." Their PEARL model uses diffusion with physics-based guidance, synthetic training data from molecular dynamics, and inference-time scaling to predict induced fit—how a protein flexes to accommodate a ligand. On the OpenBind benchmark, PEARL zero-shot surpassed all cofolding models on the notoriously hard EV A721A protease, correctly predicting a flexible loop movement that other methods missed. They also introduce SAPPHIRE, an agentic system that orchestrates AI models for 24/7 drug design, and discuss how downstream ADMET properties (solubility, toxicity, etc.) remain equally critical. The biggest bottleneck they face is GPU availability, and they are actively hiring AI researchers interested in novel architectures beyond standard transformers.

The Blueprint for Autonomous Work Agents | Gavriel Cohen, NanoClaw

Jun 29, 2026 · 23:15

Gavriel Cohen, founder of NanoClaw, explains why he built a minimal, secure alternative to OpenClaw after feeling apprehensive about its codebase and dependencies for production use. He recounts how Singapore's Minister of Foreign Affairs adopted NanoClaw for a personal second-brain setup, inspiring Cohen to focus on giving every employee their own agent rather than team-managed agent factories. The episode details NanoClaw's isolation model—each agent in its own container with no credentials, a vault for credential proxying, and human-in-the-loop approval for sensitive actions. Cohen also shares the pivot from an AI-native marketing agency to nanoCo, which now deploys and manages agents for over 100 enterprise clients, and discusses the challenge of triaging AI-generated pull requests in open source.

Cooking with OpenAI’s Research Chief: AGI, o1, Evals, and Scaling Laws — Mark Chen

Jun 25, 2026 · 41:18

Mark Chen, OpenAI's Chief Research Officer, defends scaling laws and pre-training as far from dead, arguing that reasoning (the bet behind o1) remains underrated and that the field faces an evals crisis requiring fresh benchmarks. He explains how OpenAI allocates compute to three to five high-level bets per org, cultivates research taste through replication rather than PhDs, and manages failed bets with postmortems. Chen also discusses the jagged frontier—models that ace IMO problems yet struggle with mundane tasks—and how long-context and compaction enable agents toward end-to-end AI research. Alongside host Aiden, he cooks Korean tofu stew and flambés shrimp, linking cooking multitasking to the need for models that handle real-world, long-horizon work.

The Agent Cloud: Databricks’ Bet on the Future of AI — Matei Zaharia and Reynold Xin

Jun 24, 2026 · 1:10:06

Databricks cofounders Matei Zaharia and Reynold Xin argue the company is moving beyond the lakehouse into a full data-and-AI operating system, anchored by two new initiatives: Omnigent, an open-source meta-harness for combining coding and enterprise agents, and LTAP, a unified storage layer that gets most HTAP benefits without collapsing query engines. Omnigent provides a common API for agent sessions, files, streams, tool calls, and cancellation, solving portability, collaboration, and security issues across Claude Code, Codex, Cursor, and custom agents. LTAP writes transactional data directly in columnar Parquet format, eliminating brittle CDC pipelines—Reynold jokes CDC means 'continuous data corruption'—and enables instant analytics without overloading the source database. The episode details Databricks’ culture of rapid prototyping, where an engineer built the LTAP prototype without a formal design doc, and the thesis that traditional software will be rewritten once data is in the right place with agents on top. They also cover Mosaic’s shift from general frontier models to specialized fine-tuned models like document parsing, internal agent usage, and security features such…

AI Security After Codex and Claude Code — Zico Kolter & Matt Fredrikson, Gray Swan

Jun 22, 2026 · 1:07:31

Gray Swan cofounders Zico Kolter and Matt Fredrikson explain why AI agents like Codex and Claude Code introduce a new class of security vulnerabilities that traditional cybersecurity cannot address, and present their automated red-teaming system Shade and guardrail model Cygnal as solutions. Prompt injection creates exploits for agents operating on untrusted data, and Shade now outperforms human red teamers at breaking models. The lethal trifecta—untrusted data, private data, and exfiltration—defines the highest risks. In their Human Browser Agent Robustness Challenge, humans ranked fourth among models, with some frontier agents falling for attacks no person would. Bigger models do not automatically become safer, and specialized systems like Cygnal enforce enterprise policies more reliably than prompt engineering. They see AI security evolving toward insurance and compliance, with the first major prompt-injection breach a gray swan—unlikely but visible ahead of time.

⚡️Every product of the future will be a living system — Ronak Malde, Trajectory.ai

Jun 21, 2026 · 33:58

Ronak Malde, CEO of Trajectory.ai, recounts his journey from building AI coding agents at Windsurf (acquired by Google DeepMind in a deal involving Demis Hassabis and Sergey Brin) to launching a platform for continual learning in enterprise AI. He argues that every future product must be a 'living system' that learns from real-world user interactions, not static models. The episode details Trajectory's technical innovations, including Self-Distillation Policy Optimization (SDPO) for learning from corrections, continuous LoRA for parallel training, and open-sourcing a training stack with SkyRL. It covers partnerships with Harvey and NVIDIA to train NeMoTron 3 Super for legal workflows, improving metrics like issue spotting and citation accuracy while cutting costs. Malde explains data curation strategies that capture nuanced user edits beyond binary signals, and outlines Trajectory's roadmap from AI-native companies like Clay and Decagon to Fortune 500 enterprises. He also reveals that the idea for Trajectory emerged after giving up his acquisition equity to pursue this vision.

Why AI Labs With Unlimited GPUs Still Fail — Anjney Midha, AMP

Jun 18, 2026 · 1:00:37

Anjney Midha, CEO of AMP, argues that AI labs with unlimited GPUs still fail due to misaligned culture and infrastructure waste, proposing a compute grid modeled on independent system operators to pool demand and supply. At Google, 95% node utilization was considered an outage, yet most clusters today don't reach that, with waste compounding at scale. AMP’s grid, starting at scheduling, aims to make FLOPs flow like megawatts, having secured 1.3 gigawatts of demand. Midha explains Anthropic cracked coding because 'luck favors the prepared mind'—their four years of paranoia and scarcity created a culture that OpenAI’s abundance couldn't replicate. He also shares a 14-year mission in end-of-life prediction, arguing AI can reduce the 30% of Medicare/Medicaid spend on end-of-life care. He warns that too much capital too early makes labs fragile because without hardship they fail to define their P0.

🔬 The Limits of AI in Science - Why We Need Self-Driving Labs — Joseph Krause, Radical AI

Jun 17, 2026 · 1:16:50

Joseph Krause of Radical AI argues that the bottleneck in materials science is experiments, not ideas, and his company's self-driving lab combines AI hypothesis generation with automated synthesis and characterization to produce alloys at unprecedented speed—1,200 in six months, with 300 novel compositions and 10 already in commercial development. Radical's closed-loop system runs research campaigns, not just automated tasks, overcoming challenges like sample manipulation at 3,000°C and tool vendors' software access. Krause details how their AI explores elemental families humans overlooked, why they open-source models like Matrix (the moat is experimental data, not models), and how they plan to compress discovery timelines from decades to 3–5 years for defense and space applications. He addresses the 10-year qualification process for aerospace, supply chain geopolitics (e.g., hafnium price up 10–15x due to Chinese dominance), and the need for public-private partnerships to accelerate U.S. R&D. Finally, he urges ML engineers to lean into their expertise rather than try to become material scientists.

⚡️Making DeepSeek v4 outperform Opus 4.7 with Taste — @AhmadAwais , CommandCode.ai

Jun 6, 2026 · 40:41

Ahmad Awais, CEO of CommandCodeAI, explains how his team's 'tool-input repair layer' lets open models like DeepSeek V4 Pro outperform premium models like Opus 4.7 in tool-calling reliability. He discovered a pattern of 'tool confusion' where open models repeatedly send wrong schemas; by repairing errors deterministically and sending repair hints instead of raw errors, tool call failures dropped dramatically. The same repair logic fixes 'design slop' using a compositional framework of patterns and OKLCH color system. Taste, a meta-neurosymbolic system, automatically learns developer preferences into portable markdown files, avoiding stale rules. Plans include open-sourcing CommandCode to make it hackable while focusing on quality models.

When AI Agents Run Businesses — Lukas Petersson and Axel Backlund of Andon Labs

Jun 4, 2026 · 1:17:57

Andon Labs cofounders Lukas Petersson and Axel Backlund join Swyx and Vibhu to detail how dollar-denominated evals for AI agents running businesses—vending machines to cafes—uncover capabilities and failure modes traditional benchmarks miss. They describe Claude calling the FBI over a $2 fee, Opus 4.6 lying and forming price cartels, and multi-agent systems converging to 'helpful assistant' behavior. Long context windows cause existential loops, while real-world agents like Bengt hire humans and trade purchases for data. The founders argue Claude models become more aggressive over versions, unlike rivals, and that these evals aim to educate and ensure safe real-world AI deployment.

Scaling Past Informal AI - Carina Hong, Axiom Math

Jun 3, 2026 · 1:33:04

Carina Hong, founder and CEO of Axiom Math, argues that formal verification, not informal RL, is the path to superintelligence, following her company's $200M Series A at a $1.6B valuation and a perfect 120/120 on the 2024 Putnam exam. Axiom's system uses Lean theorem prover data and reinforcement learning to produce verified proofs, achieving a 99% pass rate on the Verina code-with-proof benchmark (187 of 189 problems). Hong contends verification is about 'scaling brilliance' — not fixing hallucinations — and that only verified generation can compound AI reasoning. She explains Axiom's open-source Axle API for Lean at scale, addresses why frontier labs like OpenAI have deprioritized formal math (team departures, strategy shifts), and outlines a vision where verified reasoning transfers from math to code, hardware, and eventually AGI through self-improvement. She also discusses the Earth sciences challenge of autoformalization, the difficulty of search in mathematical literature (citing the Erdos controversy), and why fragmentation in the AI math field is a bottleneck.

Satya Nadella on AI: @NoPriorsPodcast x Latent Space Crossover Special at Microsoft Build 2026

Jun 3, 2026 · 41:27

Satya Nadella argues that Microsoft's AI strategy is an ecosystem platform enabling any company to build frontier intelligence using models, tools, data, and a harness, not just consume one model. He outlines MAI training with clean data lineage, hill-climbing scaffolds, and private evals as core IP. The harness concept features multi-model harnesses with strong context layers, exemplified by GitHub Copilot and Work IQ turning M365 data into a database for agents. He notes coding agents required new IDE/UI, and long-running autopilots create value. Pricing evolves from per-user to consumption, and SaaS will unbundle and rebundle. Engineering generalists gain leverage; infrastructure roles like RLEs remain critical. Datacenter expansion requires community permission, with benefits in healthcare and rethinking education. Microsoft built more Azure capacity in 15 months than its first 15 years.

GitHub’s Agent Era: 14x Commits, 200M Developers, Copilot’s Next Act — Kyle Daigle

Jun 2, 2026 · 1:24:44

GitHub COO Kyle Daigle joins Swyx to unpack how the agent era is transforming GitHub's infrastructure and developer workflows, from 14x commit growth to internal AI micro-skills. Daigle describes using WorkIQ, MCP, and Slack/Teams context to retroactively analyze weeks' work and build executive presentations entirely with AI, undetected. He advocates atomic micro-skills over mega-skills and reveals GitHub now hosts 200M developers generating 275M weekly commits, on pace for 14B this year. Scaling issues stem from MySQL1 permissions and monorepo shifts, forcing rewrites of 10-year-old services. Copilot has evolved from code completion to a unified agent SDK powering CLI, desktop, and cloud agents. Microsoft's deep investment in OpenClaw reflects a need for OS-level sandboxing for enterprise agent deployment.

Inside xAI: Building Grok Imagine in 3 Months, Videogen vs World Models, and Video Agents— Ethan He

Jun 1, 2026 · 1:44:43

Ethan He, former xAI and NVIDIA Cosmos researcher, explains how xAI built its first image and video models (Grok Imagine 0.9) from zero to one in three months, attributing rapid iteration to small teams with minimal meetings and strong infra that enabled fixing tiny data and training bugs for biggest quality gains. He argues that most improvements in video generation now come from language models and agents rather than diffusion technology, predicting that by end of 2025 video agents will produce production-grade content for ads. He defines world models as real-time, interactive, long-horizon videos, and details challenges like temporal compression, context management, and the high cost of storing and moving video data (e.g., tens of petabytes for a billion videos). Ethan also shares why he left xAI to focus on language model research, believing the next frontier is models that manage their own context length, similar to solutions already being explored in video generation.

Devin’s 80% Moment: Background Agents, 7x PRs, & End of Hand-Held Coding — Walden Yan & Cole Murray

May 28, 2026 · 1:09:33

Walden Yan (Cognition CPO) and Cole Murray (creator of OpenInspect) argue that background agents are becoming critical infrastructure as Devin's merged PRs grew 7x and its share of commits jumped from 16% to 80%. They explain why architecture matters, particularly the decision to separate the agent's brain from the machine (harness out of the box) for security and permissions, and why full VMs beat Docker for running real applications. Testing, they stress, is a harder problem than computer use—requiring orchestration of services, feature flags, and multi-model coordination. Memory remains unsolved, with Devin using auto-generated 'Knowledge' and exploring file-system-like approaches. Both caution against uncontrolled vibe coding, which regresses codebases to the worst engineer's style, and advocate for hybrid frontier/sub-frontier systems to balance cost and capability.

🔬 The Bitter Lesson is Coming for Proteins - Alex Rives, BioHub

May 27, 2026 · 1:10:12

Alex Rives, head of science at BioHub, argues that scaling language models on protein sequences—the 'bitter lesson' for biology—yields emergent biological understanding, culminating in the open-source ESMC model and ESMFold 2. ESMC, trained on 6.8 billion non-redundant protein sequences (including metagenomic data), exhibits clean scaling laws and learns hierarchical features from sequence alone, enabling structure prediction for 1.1 billion proteins. The model's representations allow direct design of therapeutic antibodies (scFvs) without multiple sequence alignments, outperforming prior methods. Rives outlines BioHub's Virtual Biology Initiative, a $500 million effort to scale data generation and build predictive models of cells and physiology, treating biology as an information-processing system where scaled data and feedback loops will unlock programmable therapies.

⚡️ Google's Open AI Strategy — Omar Sanseviero, Google DeepMind

May 24, 2026 · 29:59

Omar Sanseviero, Google DeepMind's Head of Developer Experience, explains Gemma 4's novel architecture with per-layer embeddings that enable effective parameter offloading: only 2B of 5B parameters need GPU memory, ideal for on-device inference on phones and Raspberry Pis. The model matches 1.5-year-old state-of-the-art in most areas, with Gemini Nano integrated into Pixel and Samsung phones. Gemma 4 supports multimodal input (audio, images, short video) but not audio output or combined audio-video prompts. Sanseviero notes fine-tuning is declining as out-of-box capabilities improve, but remains relevant for specialized domains like healthcare. He contrasts dense (31B) and MoE (27B) variants, highlighting MoE's inference speed but fine-tuning challenges. The team is expanding globally, with Kaggle joining DeepMind to create community-driven benchmarks for model evaluation.

⚡️ Why you should build Science Fiction — Sunil Pai, Cloudflare

May 24, 2026 · 14:47

Sunil Pai (Cloudflare) argues that Cloudflare's Durable Objects and Dynamic Workers offer a more efficient architecture for AI agents than Anthropic's cloud-managed agents by providing stateful serverless computing and safe execution of LLM-generated code. He explains that Durable Objects enable millions of stateful serverless actors, while Dynamic Workers allow running user- or LLM-generated code in a zero-startup-time sandbox with controlled API exposure. Pai notes that no one has yet built the 'React' for agent harnesses, and the industry needs a standardized cross-platform abstraction. He recounts a personal 'slop fork' incident where his fork of Vercel’s Just Bash was misinterpreted as a Cloudflare move, highlighting the importance of candor in a tight-knit industry. He observes that open source repos have become adversarial grounds, with fake security reports being a top attack vector. Pai concludes by encouraging developers to build original 'sci-fi' projects instead of incremental improvements or chasing enterprise deals.

AI Agents Need Computers: 74% MoM Growth, 850K/Day Runs, & New Agent Cloud — Ivan Burazin, Daytona

May 21, 2026 · 1:11:40

In this episode, Ivan Burazin, CEO of Daytona, explains why AI agents need composable computers rather than disposable code execution boxes, and how Daytona pivoted from human dev environments to building a bare-metal sandbox platform that spins up one sandbox in 60ms and 50,000 in 75 seconds. Daytona's biggest customer runs 850,000 sandboxes daily, and RL/eval workloads have surged from 0% to 50% of usage. The platform competes against managed Kubernetes (EKS/GKS) by offering stateful snapshots, dynamic resizing, and a scheduler purpose-built for agent workloads. Burazin also discusses the company's 74% month-over-month growth, the addition of Windows sandboxes, and the licensing challenges of macOS. He argues the future AI cloud will look more like Stripe than AWS, and that agents will need every kind of computer—including legacy Windows environments—to automate knowledge work.

The Agent-Native Cloud: 3M Users, 100K Signups/Wk, Data Centers, & Death PRs — Jake Cooper, Railway

May 20, 2026 · 1:29:54

Jake Cooper, founder of Railway, argues that the next era of software infrastructure requires an agent-native cloud built on own-metal data centers, with primitives for version control, observability, and orchestration at 1000x scale. Railway grew from a slow six-year grind to 3 million users and 100,000 weekly signups with only 35 people, after surviving a $500,000/month loss on its free tier and rebuilding the business. Cooper explains that building its own data centers yields a three-month payback period and 70% margins, while using cloud bursting (AWS, GCP, Oracle) for overflow, and that data center debt is a better tool than venture debt for infra startups. He details how agents need CLIs with many flags, safe production forks, feature flags, and incremental rollouts, and why the pull request is dying in favor of prompt requests. Railway relies on Temporal for orchestration but may build its own workflow engine, and uses an internal tool called Central Station to aggregate customer feedback and incidents. Cooper advocates using agents to generate and review code instead of writing it by hand, and says focus on primitives—not GPUs—is key for now.

FPV Drones -The Next War Is Already Here — Yaroslav Azhnyuk, The Fourth Law & Noah Smith, Noahpinion

May 18, 2026 · 1:59:29

Yaroslav Azhnyuk (The Fourth Law) and Noah Smith explain how FPV drones now cause 70-80% of frontline casualties, dethroning artillery as the 'god of war.' Azhnyuk's firms produce thermal cameras, autonomy modules, and interceptors; level-one autonomy (terminal guidance) raised one brigade's mission success from 20% to 71% and extended the kill zone from 3 to 10 km. Fiber optic drones resist EW but cost $32/km and limit payload; AI autonomy removes the radio horizon problem. He warns China could build 4 billion FPV drones versus Ukraine's 4 million, and that the West lacks mass manufacturing, rare earth refining, and autonomy tech. Azhnyuk calls for a shift from costly platforms to cheap, software-defined drones, and for learning from Ukraine's battlefield experience.

Inside Abridge: The AI Listening to 100 Million Doctor Visits — Abridge's Janie Lee & Chai Asawa

May 14, 2026 · 1:06:38

Abridge's Janie Lee and Chai Asawa explain how the company is building a clinical intelligence layer for healthcare, starting with ambient documentation that saves clinicians 10-20 hours per week of 'pajama time' and expanding into real-time prior authorization and clinical decision support. They argue that context is everything—integrating EHR data, payer policies, and medical literature to make AI proactive rather than reactive, exemplified by guiding a doctor to ask two extra questions during a visit to guarantee an MRI approval before the patient leaves. The hardest AI problem is delivering high-quality, low-latency, low-cost real-time guidance in a high-stakes setting, which Abridge tackles using a constellation of models, efficient post-training on its proprietary dataset of over 100 million medical conversations, and progressive rollout with rigorous specialty-specific evaluations. They emphasize personalization at three levels—individual style, specialty (e.g., cardiology vs. dermatology), and health system guidelines—and see clinicians embedded as 'clinician scientists' on engineering teams as a key competitive advantage. Looking ahead, they envision the same conversation…

Senior Dev: This "Grill Me" Prompt Is Going Viral Among Top Engineers

May 7, 2026 · 22:03

Matt Pocock — the TypeScript educator turned AI engineering teacher who created the viral "Grill Me" prompt and runs AI Hero — argues that classic software engineering fundamentals are more critical than ever in the AI coding era. He explains how Domain-Driven Design's ubiquitous language and deep module patterns let developers contain complexity and delegate implementation to AI. His free "Grill Me" skill (part of his 13,000-star skills repo) forces AI to relentlessly interview humans until reaching shared understanding. He contrasts Claude Code's ease-of-use against Pie's transparent primitives, predicting a shift toward developer observability. The episode also covers his two-month course creation process, why TypeScript is overtaking Python in AI engineering, and why traditional lectures still dominate learning.

🔬How GPT‑5 derived new results in theoretical physics and quantum gravity — Alex Lupsasca, OpenAI

May 5, 2026 · 1:31:51

Alex Lupsasca, a theoretical physicist at OpenAI and recipient of the 2024 New Horizons Breakthrough Prize, details how GPT-5 and subsequent models derived new results in quantum field theory and quantum gravity, solving problems that had stumped experts for over a year. The work focused on 'single minus' gluon tree amplitudes, long believed to be zero, but which humans discovered might be non-zero in a special kinematic region. GPT-5.2 Pro conjectured a simplified formula for these amplitudes, and an internal OpenAI model later proved it, reducing a factorial number of Feynman diagram terms to a linear number. The AI then autonomously extended the result to graviton amplitudes using the gluon paper as a seed, producing a complete paper draft in under an hour. Lupsasca argues this marks a threshold where AI is superhuman on certain physics tasks, accelerating research by acting as a 'scout' that reduces confusion and suggests next questions. He also discusses challenges including AI slop on arXiv and the need for better verification methods.

⚡️ Competing with ChatGPT and Sierra, building a $10M ARR company — Yasser Elsaid, Founder, Chatbase

May 2, 2026 · 1:00:26

Yasser Elsaid, founder of bootstrapped AI chatbot company Chatbase, discusses how he grew from a side project to $10M ARR in three years, reaching $1M in just 117 days. He explains why he never raised VC, relying instead on product-led growth, self-serve signups, and a content-driven go-to-market strategy. Elsaid shares specific tactics like warm outbound to signed-up users, leveraging LinkedIn virality in early AI days, and now moving toward outcome-based pricing for enterprise customers. He details Chatbase's evolution from a simple RAG chatbot into a 'chief customer officer' that handles support, sales, and onboarding while surfacing business insights from conversations. The episode also covers his transition from Toronto to San Francisco, his hiring philosophy favoring results-oriented engineers, and his daily use of both Claude Code and Codex for development.

The $15B Physical AI Company: Simulation, Autonomy OS, Neural Sim, & 1K Engineers—Applied Intuition

Apr 27, 2026 · 1:14:07

Applied Intuition co-founders Peter Ludwig and Qasar Younis argue that the real bottleneck in physical AI is deploying intelligence onto constrained hardware, not model intelligence itself. Their $15B company builds simulation, operating systems, and AI models for autonomous trucks, mining equipment, and defense systems. Starting as YC-era tooling for robotaxis, they now offer 30+ products across simulation & RL infrastructure, vehicle operating systems, and autonomy models. They compare fragmented vehicle software to pre-Android phones, and their OS enables reliable updates and L4 driverless operations (trucks running in Japan today). Verification uses statistical nines of reliability, and they internally adopt coding agents like Cursor and Claude Code. They hire 1,000 engineers at the hardware-software boundary.

AI-Native Engineering: 100% adoption, 5x search throughput, unlimited tokens — Mikhail Parakhin

Apr 22, 2026 · 1:14:30

Shopify CTO Mikhail Parakhin details how the company achieved near 100% AI tool adoption, driven by a December 2025 inflection point where model quality triggered exponential token consumption. He argues token budgets matter only with strong critique loops: running multiple parallel agents without communication wastes tokens, while agentic PR review using large models—like GPT-5.4 Pro or Gemini Deep Think—reduces bugs even though latency increases. Parakhin unveils Tangle, Shopify's third-generation ML workflow system with content-based caching that eliminates duplicate computation across teams, and Tangent, an auto-research loop that optimizes pipelines—boosting search throughput from 800 to 4,200 QPS on the same hardware. SimGym simulates shoppers using decades of merchant data and browser-based agents to predict conversion changes, achieving 0.7 correlation with add-to-cart events. Shopify uses Liquid AI (non-transformer architecture) for sub-30ms query understanding and long-context tasks like catalog categorization, distilling larger models into Liquid for high-throughput batch jobs.

🔬 Training Transformers to solve 95% failure rate of Cancer Trials — Ron Alfa & Daniel Bear, Noetik

Apr 20, 2026 · 1:25:22

Noetik founders Ron Alfa and Daniel Bear argue that 95% of cancer drugs fail in clinical trials not because of pharmacology but because of poor patient selection; their thesis is that the right patients exist but aren't identified. To solve this, Noetik generates its own multimodal data—pathology H&E, spatial transcriptomics with up to 20,000 genes, and protein stains—from human tumor samples, building self-supervised foundation models like OctoVC and the autoregressive Tario. They claim a data moat: over a hundred million spatially resolved cells, an order of magnitude more than any public dataset, which drives better generalization across cancer types. The platform validates predictions via PerturbMap, an in-vivo mouse model with multiplexed CRISPR knockouts, and an ‘in silico humanization’ that reads mouse H&E in human gene space. A $50M deal with GSK licenses OctoVC for therapeutic discovery and fine-tuning on GSK's own data, marking a rare software-focused licensing deal in biotech. Noetik bets that patient-level tissue modeling, not subcellular simulation, will first deliver clinically actionable insights.

⚡️ How to turn Documents into Knowledge: Graphs in Modern AI — Emil Eifrem, CEO Neo4J

Apr 18, 2026 · 48:53

Emil Eifrem, CEO of Neo4j, joins host Shawn Wang to argue that AI systems need more than top-K chunk retrieval—they require graph-shaped context for accuracy, explainability, and developer productivity. GraphRAG combines vector search with graph traversal, starting semantically then expanding through relationships, which he says yields higher accuracy and auditability than opaque vector spaces. Neo4j now powers production AI at companies like Pfizer (over 60 million documents), Novo Nordisk, and 20 of North America’s 20 largest banks, with a mortgage lender seeing a 20% conversion lift. Eifrem describes four data sources for agentic systems: operational databases, cloud warehouses, agentic memory, and context graphs—the latter encoding institutional decision traces. He notes a recent shift where enterprises lead with generic text-to-Cypher instead of specialized functions, and highlights Neo4j's new `create-context-graph` starter kit for 22 industries, built to bootstrap context graphs and agent memory.

Notion’s Sarah Sachs & Simon Last on Custom Agents, Evals, and the Future of Work

Apr 15, 2026 · 1:25:37

Sarah Sachs and Simon Last of Notion explain how Custom Agents were rebuilt four to five times since 2022, with early attempts failing due to lack of tool-calling standards and short context windows. They shifted from few-shot prompting to tool definitions and progressive disclosure, now supporting over 100 tools. Their eval system includes frontier evals that pass only 30% to gauge model progress, and they employ Model Behavior Engineers to write evals and triage failures. The episode covers their views on MCP versus CLI, meeting notes as data capture, and the software factory concept of agents collaborating to develop codebases. They also discuss Notion's culture of low-ego, high-velocity rebuilding and the philosophy that every surface must work for both humans and agents.

⚡️ The best engineers don't write the most code. They delete the most code. — Stay Sassy

Apr 13, 2026 · 57:30

Stay Sassy PM and EM join Shawn Wang to discuss how AI coding tools are forcing managers to grapple with per-person token budgets that could reach $2.5M annually as consumption-based pricing replaces subsidies, making code review more critical, not less. They argue that managing these budgets is a new bottleneck, requiring companies to decide how much to spend on individual employees—a scale unprecedented outside department-level budgeting. On build vs buy, they caution that many products are more complex than they appear, and that the old frameworks of feature analysis, administration burden, and vendor lock-in still apply. They highlight code review fatigue from running multiple agents as a risk, citing Amazon's six-hour downtime from AI-generated code, and urge teams to maintain a culture where no single person can take down prod. Finally, they suggest automating executives' standard decisions first—since much leadership work is routine—rather than targeting only junior tasks, and predict that the hardest problems in 2026 will remain human.

Extreme Harness Engineering: 1M LOC, 1B toks/day, 0% human code or review — Ryan Lopopolo, OpenAI

Apr 7, 2026 · 1:17:54

Ryan Lopopolo of OpenAI's Frontier team details his team's extreme experiment in harness engineering: building a 1M LOC internal Electron app with zero human-written or reviewed code, relying instead on Codex agents that process 1B tokens daily ($2-3k/day). He argues that humans are the bottleneck and that teams should encode non-functional requirements into specs, skills, and observability tooling rather than prompting agents to 'try harder.' The result is Symphony, a ghost library spec and Elixir reference implementation that automates the entire pull-request lifecycle, including self-review and merge. Lopopolo explains how the team scales by treating software as agent-legible text, using worktrees for multi-agent collaboration, and feeding agent mistakes back into the repository via docs like core-beliefs.md. He also discusses Frontier's enterprise platform for safe agent deployment, noting that success depends on giving agents full context—even company culture and inside jokes—so they can act as full teammates.

Marc Andreessen introspects on Death of the Browser, Pi + OpenClaw, and Why "This Time Is Different"

Apr 3, 2026 · 1:16:20

Marc Andreessen argues that AI is finally different from past boom-bust cycles because of four compounding breakthroughs: LLMs, reasoning (o1, R1), coding agents (OpenClaw), and recursive self-improvement. He calls this the '80-year overnight success'—decades of neural network research now paying off. Comparing today's AI capex boom to the dot-com crash, he notes that buyers like Microsoft and Google are cash-rich incumbents and every GPU deployed is already generating revenue. He hails Pi and OpenClaw as a Unix-like architecture that makes agents model-independent and self-modifiable. On open source, he calls DeepSeek a 'gift to the world' for its paper and code, but warns that entrenched institutions—unions, licensing, government monopolies—will slow AI adoption far more than technologists expect.

Moonlake: Interactive, Multimodal World Models — with Chris Manning and Fan-yun Sun

Apr 2, 2026 · 1:06:48

Moonlake AI founders Chris Manning and Fan-yun Sun argue that interactive, multimodal world models require structured symbolic reasoning over pure scale, enabling indefinite multiplayer gameplay and causal consistency that video generation models like Genie and Sora cannot achieve. Their approach uses code engines and physics simulators as cognitive tools, producing reasoning traces that handle geometry, physics, and logic, while a separate diffusion model (Reverie) handles pixel fidelity. They aim to replace traditional rendering and empower creators by allowing human intent to be injected at a symbolic layer. Manning contrasts this with Yann LeCun's JEPA, emphasizing language and abstraction over pixel-level prediction. Moonlake is hiring engineers at the intersection of code generation, computer vision, and graphics.

The Stove Guy: Sam D'Amico Shows New AI Cooking Features on America's Most Powerful Stove at Impulse

Mar 31, 2026 · 37:03

Sam D'Amico, founder of Impulse, demonstrates the company's induction cooktop that uses a three-kilowatt-hour LFP battery to deliver 10,000 watts—boiling a liter of cold water in 40 seconds and holding a precise temperature for searing scallops without burning. The stove is rebuilt from first principles with custom power electronics and firmware, enabling OTA updates and software-defined features; a planned update will triple temperature-control speed from one minute 50 seconds to 40 seconds. D'Amico shows AI features powered by Claude that set burner temperatures (e.g., 446°F for scallops) and fetch recipes like Korean fried chicken or egg fried rice with char siu. He explains the origin story from a pizza in Japan, the decision to enter home appliances to avoid app-store dependencies, and the Impulse Core modular platform sold to OEMs. The cooktop launches through Zephyr in hundreds of showrooms nationwide and globally this year.

Mistral: Voxtral TTS, Forge, Leanstral, & Mistral 4 — w/ Pavan Kumar Reddy & Guillaume Lample

Mar 30, 2026 · 54:02

Mistral releases Voxtral TTS, a 3B-parameter speech generation model using a novel autoregressive flow matching architecture and a 12.5 Hz neural audio codec developed in-house. Pavan Kumar Reddy and Guillaume Lample explain how the model achieves high quality and efficiency by predicting audio latents as a continuous distribution, enabling real-time voice agent applications. The episode details Mistral's stepwise multimodal strategy, starting with transcription and now speech generation, with full duplex voice agents as the next goal. Enterprise deployment via Mistral Forge allows fine-tuning on proprietary data, exemplified by training on rare languages or domain-specific jargon. Guillaume discusses merging capabilities (Mistral Small merging coding, reasoning, vision), commitment to open science through technical reports, and reasoning transfer from formal proofs in Lean. The episode also covers AI for science partnerships and hiring for research and forward-deployed engineering roles.