PodcastsHear the voice. See the shape of the thought.

GitHub COO Kyle Daigle joins swyx to map what the agent era looks like from inside the platform hosting 200 million developers and now processing commits at 14x last year's pace. Across 84 minutes they cover how Kyle runs GitHub with AI-driven micro-skills and WorkIQ MCP, why former developers in leadership have an unusual edge right now, the full arc of GitHub's platform history from webhooks to Actions to Copilot, and where trust in agent-generated code ultimately has to come from. The conversation is grounded throughout in Kyle's own weekend and executive workflows: building AI-generated revenue presentations, running 15 simultaneous agents on a Saturday, and describing what "ambient AI" would actually need to do before it becomes genuinely useful. ## [00:00] Hook Kyle opens mid-sentence, already deep in his argument: people who detoured into other careers before coding, and came back armed with cross-domain knowledge, are uniquely positioned in the AI era. Running 15 agents on a Saturday while his kids are at lacrosse is not just a productivity flex — it recreates the feeling of creation that got him into software in the first place. > *"I can crank up 15 agents on Saturday, you know, while my kids are doing lacrosse. That's like really powerful and I think it gets me back to that feeling of like creation."* ## [01:21] Introduction Kyle's title is COO of GitHub, but he recently took on CMO of Developer for Microsoft as well — meaning every developer-facing product and communication across the broader Microsoft ecosystem now runs through him. He's been at GitHub for 13 years, joined as a developer, personally built webhooks and the platform/API layer, ran engineering until 2018, then moved into the operational and business side. The dual COO/CMO role is unusual; Kyle frames it as the same job with a larger surface area: tell the truth, be authentic, let the products speak. > *"I built webhooks and worked with teams building the API, built the platform layer, anything that integrated with GitHub, up until really 2018 I built or ran the engineering teams."* ## [04:57] Why AI Got Kyle Coding Again Swyx points out that Kyle's commit graph shows a clear dip through his leadership years and a sharp uptick recently — entirely driven by AI. Kyle is not writing features for GitHub's product; he's building internal agents and workflow tools that stitch together disparate data sources. His primary use case is retrospective: using WorkIQ, MCP servers, Slack, Teams transcripts, and Obsidian notes to ask "what actually happened last week, what worked, and what should I tweak for the next few days." He finds LLMs are exceptionally good at pattern-finding across a week of context, far more so than generating forward-looking plans from scratch. > *"I find AI in like what most of this launch here is actually like less building forward. It's actually like a recursive loop backwards. I'm always looking at what had happened first."* ## [08:25] Running GitHub with AI: WorkIQ, MCP, Slack, Teams, and Skills GitHub rolled out AI internally by meeting people where they already work — Slack, Teams, email — rather than forcing them onto a new tool. Every employee, technical or not, gets the Copilot CLI plus a shared set of atomic micro-skills deposited into repos. The era of the "mega-skill" that handles an entire workflow end-to-end is over; what works are tiny, single-purpose skills that do one thing well and compose cleanly. Kyle uses Postel's Law as a design principle: liberal in what each skill accepts, strict in what it outputs. WorkIQ, the M365 MCP server, lets anyone ask backward-facing questions across every meeting, email, and chat — critical for a fully remote, globally distributed team. > *"We're ending the era of these like massive beautiful perfect skills. What we found is these incredibly micro skills that are just doing one thing for us very very well versus a skill that's going to do that full report that doesn't really exist on our side anymore."* ## [17:00] The Golden Age for Former Developers in Leadership Swyx asks whether people like Kyle — technical backgrounds, now in exec roles — have a structural advantage in the AI era. Kyle's answer: pattern-finding and problem-solving are the durable skills from his developer years, and AI has given him back the ability to apply them directly in code. The more interesting case isn't developers going back to update old side projects; it's people who spent ten-plus years accumulating business knowledge now using that context as leverage when wielding AI tools. The cross-domain background, once a liability in pure engineering orgs, is now a multiplier. > *"I just find that the folks that came from a different career, went to school for something else, went off and did this random thing and then became a software dev — now having the power of an AI where I can crank up 15 agents on Saturday."* ## [18:52] 15 Agents on Saturday and AI-Generated Executive Work Kyle built GitHub's annual revenue planning presentation entirely with AI — a SQLite app to view the data, skills pulling from Obsidian notes and work context, and a deliberate skill that made the output look "humanly bad" so it wouldn't read as AI-generated. He presented it to the CRO and CFO teams without disclosing the process; nobody asked. His point isn't to hide AI from colleagues but to demonstrate that value is in crafting and judgment, not slide assembly. The ability to build a small data-manipulation app and control the final output is, specifically, the advantage that developers carry into leadership. > *"I ultimately built this entire presentation without touching any of it. And I was like, okay, I'm just going to present this to our CRO, the CFO, their teams without mentioning I built it with AI. Never came up once."* ## [21:41] How AI Changes the Chief of Staff Role Kyle still has a chief of staff — but the job has shifted. Slide prep and presentation assembly have moved to AI; what remains irreplaceable is the human connective tissue: knowing which people in which cities should meet, surfacing relationship opportunities across a distributed org, brokering conversations that don't appear in any MCP server. The analogy is email replacing letter-opening: nobody expects the chief of staff to open physical mail anymore, and soon nobody will expect them to build decks either. The judgment about *who* should talk to *whom* is what stays. > *"I still have a chief of staff because the difference is the human connection aspects — I should be meeting with this group and this team and they have an opportunity and I'm going to be in San Francisco today."* ## [23:06] GitHub's History: Actions, npm, Webhooks, and Open Source Kyle walked the platform's architectural history: GitHub Services (pre-2014 arbitrary Ruby execution with no real containerization), webhooks, Pages, and then Actions — launched by Kyle personally at GitHub Universe in October 2018. Actions went from "we should not be running arbitrary Ruby on people's behalf" to a fully containerized compute layer now using Azure Dev Compute for fast, small-VM agent spin-ups. The npm acquisition came from a simple premise: npm was powering the internet and having scaling problems; GitHub's job was to keep it running and raise its security posture. Every security improvement — 2FA enforcement, token invalidation on exposure — breaks something downstream, and that balance between hardening a 15-year-old ecosystem and not causing developer snow days remains the central tension. > *"We have changed the 2FA policies, we've changed the way the tokens work. When we find tokens that have been exposed or potentially exposed, we invalidate them. That creates issues. But we're trying to push the community forward."* ## [30:06] Slop Forks, Vendoring, and AI Dependency Management Swyx raises the "slop fork" pattern — AI-assisted vendoring where you pull in only the source you need rather than importing a whole package — and asks whether it sidesteps npm's vulnerability surface. Kyle: vendoring was how everyone worked in 2013, and there's something true about pulling in only what you need, but it doesn't fix the fundamental problem. An agent evaluating code can be convinced it's secure just as easily as a human can. Static analysis and runtime testing still need investment regardless of package scope. GitHub's historical stance — wait for community RFC and social consensus before cementing a practice — means they won't push a single vendoring standard, but will build tools for maintainers to enforce their own trust rules. > *"The vulnerabilities — in an agent looking at them there's time and time again a million different ways in which we can convince an agent that this thing is like secure or not."* ## [35:18] Pull Requests, Prompt Requests, and Trust in Agent-Generated Code GitHub invented the pull request as a social trust mechanism, and now agents are generating the majority of PRs on many projects. Kyle assessed various alternatives — Peter Coppola's "prompt request" model, Thomas Dohmke's contribution-asset approach — but argues that none fully solve the underlying problem: trust is social, not technical. Even if a PR is 100% verified by static analysis, humans still reach for human signals (does Mitchell approve it?) before merging. GitHub's current direction centers on giving maintainers malleable tools to define their own trust heuristics rather than imposing a universal standard, because any single standard immediately becomes a gamification target. The endgame is something closer to human digital identity. > *"The reason why there's not a single answer is ultimately we're trying to codify trust. Right now when an agent writes code and another agent reviews code and then Kyle goes and looks at it, the trust is kind of diffuse."* ## [42:42] GitHub Stars, 200M+ Developers, and the New AI Builder Wave GitHub crossed 200 million accounts — up from 80 million not long ago. The rapid star accumulation on new AI projects is mostly genuine: an entire new cohort who built their first app in the AI era is swarming the zeitgeist. Kyle refuses to split hairs about who "counts" as a developer, drawing on his own experience being called a fraud for having a GitHub account before he knew what git was. The gamification problem is real (whack-a-mole anti-abuse, now AI-powered), but the majority of the star velocity is new builders who want to participate in the moment the way Kyle wanted to participate in the Ruby era. > *"It's not just developers. It's folks that have maybe started coding or only joined in since the AI era. And those projects are going up because you want to be a part of this moment."* ## [46:36] GitHub Spark, Low-Code, and Why GitHub Still Shows the Code GitHub experimented with Spark as an easy app-build-and-run experience. The lesson: for developers, the value was always simple runtime, not a UI veneer hiding the code. GitHub's architectural principle is non-negotiable — they will always show you the code. The broader goal Kyle articulates is lowering the barrier to that first "I had an idea and I built it" moment: anyone should be able to swap a light switch without needing to open the breaker box. > *"Anytime we try to put a veneer on top of something, we still always show you the code. That's kind of like a tenant. We're never gonna hide the code from you ever."* ## [48:59] GitHub's Hardest Era: 14x Growth, Reliability, and Scale GitHub went from 1 billion commits in all of 2025 to 275 million per week in April 2026 — a 14x year-on-year rate still accelerating. This broke things in new ways: not the old webhooks reliability problems (those were fixed and rewrote), but novel permission-layer failures only visible at cross-object scale. The core pain point is MySQL 1, a monolithic permissions database GitHub has been decomposing for years; permissioning is where most cross-cutting outages originate. Simultaneously, the industry is shifting back toward monorepos, which carry unique git infrastructure performance characteristics. Kyle frames the scaling problem as "diagonal" — vertical and horizontal both stop working, so you crack open services running unchanged for 10-15 years and rewrite them. > *"We're doing more in a month than we did in a year last year. By roughly every measure, there's growth that is much much bigger. And that is breaking our system in new ways, not old ways."* ## [60:42] Actions as the Compute Layer for CI/CD and Automation Actions has evolved well beyond CI/CD into a general-purpose automation compute layer — the root of significant availability pressure because every agent task and agentic workflow translates into more builds and more CPU. GitHub is expanding compute through both its own data centers and Azure cloud, and is using Azure Dev Compute (fast small-VM spin-up) under the hood for containerized agent execution. The path to fewer outages is a step-change model: large foundational infrastructure fixes that take time, then visible plateau improvements in availability rather than incremental noise reduction. > *"Actions is the core compute layer for either CI or side project. More tools, more agents, more PRs mean more builds. More builds need more CPUs and we simply need more CPUs."* ## [63:25] The State and Future of GitHub Copilot Copilot's history: launched as code completion, then shifted energy toward fine-tuning as the industry demanded better accuracy, and then next-gen models arrived and made fine-tuning less critical — creating confusion about where Copilot was going. The current architecture unifies a single SDK and agent harness across code completion, the new CLI, the new desktop app, and cloud agents. The future Kyle describes covers the full SDLC: security remediation, issue triage, documentation drift detection — not just writing code. The remaining hard problem is context and memory: getting GitHub to "act like Kyle wants it to act" across all his dependencies, preferences, and team context. > *"What we think is that it's not solely about the code generation. It's really about having the ability to use these coding agent brained harnesses across not just the coding experience but also security remediation, every GitHub issue that comes in."* ## [69:45] Ambient AI, Background Agents, and the Future of the SDLC Kyle argues the industry is still stuck in a "hyper-myopic" frame where coding agents only know about code. What he actually wants is ambient AI that carries every spec doc, every email thread, every conversation, every Obsidian note into its decision-making as a developer — not as a recall tool you query, but as persistent background context that shapes implementation choices in real time. OpenClaw interests him precisely because it connects personal context to agent action; but the missing piece is making that context available *during* software development. The extreme version — AI that proactively directs you rather than waiting to be asked — is the inversion of control that both excites and slightly alarms him. > *"The most interesting thing to me in AI is actual ambient AI. I'm looking to be implementing a new feature and for it to know every spec doc, every email, the conversations that I've had online, everything about how this could be implemented and be able to use that as part of its decision-making."* ## [74:30] OpenClaw, Enterprise Security, and the New OS for Agents Microsoft has a CVP dedicated to OpenClaw — unusual given Microsoft doesn't own Anthropic. Kyle explains: OpenClaw demonstrated what a valuable personal agent actually looks like (full personal context, computer use, not just chat), and Microsoft's job is to make that work in enterprise — OS-level sandboxing on Windows so you can run an agent on a work device without it becoming a security incident. The framing Kyle reaches for: Microsoft is the original operating systems company, and agents need a new OS layer. Workloads have changed so fundamentally that the right question is no longer "do we need more inference?" but "what type of compute do we need to run these agentic flows?" — all the way down to silicon. > *"Microsoft is the original operating systems company and here's the new operating system for AI. Operating systems need to look different than they looked five years ago because it's not just you using them anymore."* ## [79:24] Build Announcements, WorkIQ, FoundryIQ, and Microsoft Context Kyle previews what GitHub and Microsoft are announcing at Build: WorkIQ (M365 context engine via MCP, powerful for retrospective questioning across all work assets) and FoundryIQ (same intelligence layer that connects to existing data stores without requiring migration). The pitch for enterprise developers: "how I build on the weekend should be how I build at work" — but Fortune 500 companies can't just vibe-code and ship; security and compliance gates have to move as fast as development does. WorkIQ and FoundryIQ are the attempt to bring weekend-level agility into the enterprise context layer, with the governance that lets it survive in large organizations. > *"Work IQ, Foundry IQ — these context engines are wild good and we've given them to our developers at GitHub. You can ask questions around everything in your work context and it's surprisingly powerful."* ## [83:02] What Should swyx Ask Satya? swyx is about to interview Satya Nadella at Build and asks Kyle what to ask. Kyle's recommendation: challenge Satya on what he believes is demonstrably true about the AI and inference landscape in two to three years — not as a throwaway futurist question, but as a direct test of the internal bets Microsoft is making right now. Significant external skepticism exists about Microsoft's AI approach, and a straight answer from Satya would be both a genuine stress test and a reassuring signal for the developer community. > *"The best question to ask is what he thinks is true in like two or three years from now. The way that he is looking at this AI problem, the inference problem, the token problem — why is this approach in two years going to pay off?"* ## Entities - **Kyle Daigle** (Person): COO of GitHub and CMO of Developer for Microsoft; 13-year GitHub veteran who built the original webhooks and platform API layer. - **swyx** (Person): Host of Latent Space podcast; developer-advocate-turned-podcaster who conducted this interview at Microsoft Build 2026. - **GitHub Copilot** (Software): GitHub's AI coding assistant, now spanning code completion, CLI, desktop app, and cloud agents under a unified SDK. - **WorkIQ** (Software): Microsoft 365 MCP server that gives employees a context engine over all work assets (Teams, email, calendar, etc.). - **FoundryIQ** (Software): M365 intelligence layer that connects to existing enterprise data stores without requiring migration. - **GitHub Actions** (Software): GitHub's general-purpose compute and CI/CD automation layer; primary source of CPU demand growth from agent workloads. - **OpenClaw** (Software): Anthropic's Claude Code agentic tool; referenced as a model for what a personal AI agent with full context and computer use looks like. - **npm** (Software): JavaScript package registry acquired by GitHub; central to supply-chain security discussions about vendoring, slop forks, and dependency trust. - **Mitch Hashimoto** (Person): Co-founder of HashiCorp, active open-source maintainer; discussed in context of vendoring approaches and GitHub's maintainer relationship model. - **Thomas Dohmke** (Person): CEO of GitHub; referenced in context of PR workflow evolution. - **Microsoft Build** (Organization): Annual Microsoft developer conference; context for this episode's release and Kyle's expanded-role announcements.

Ethan He built NVIDIA's Cosmos world model, then joined xAI mid-2025 to build Grok Imagine from scratch — no infra, no data, no model — and shipped the first audio-video generation model in three months. He walks swyx and Vibhu through the full technical stack: synthetic captioning pipelines, VAE design tradeoffs, step distillation, audio-video alignment, and the hard economics of storing petabytes of video training data. His central argument runs through the entire conversation: since diffusion model technology has largely matured, most quality gains in video now come from language models, not from the video model itself — a view with direct implications for where the field goes next, including video agents, generative UI, and embodied world models. ## [00:00] Hook This exchange — Ethan's "pretty big claim" that visual intelligence now mostly comes from language — is pulled from later in the interview, where he argues that improvements to video models are increasingly driven by better language models acting as prompt rewriters and orchestrators, not by advances in diffusion or flow-matching architectures themselves. > *"Every time you see there's some improvement on these models, I would say mostly the gain comes from language model, not coming from the video model itself."* ## [01:16] Introduction swyx and Vibhu welcome Ethan to the Latent Space studio, noting he has been a recurring presence through the podcast's paper club — first presenting the Cosmos world model paper, then mixture-of-experts work. The conversation opens with a brief aside about the Poolside paper released the same day, a fully open Gemma-level model trained on 40 trillion tokens, before pivoting to Ethan's own trajectory. ## [02:41] From NVIDIA Cosmos to xAI Ethan built Cosmos — NVIDIA's giant video foundation model aimed at giving roboticists a simulatable world to build on — and shipped it by end of 2024. Once he realized video models obeyed the same scaling laws as language models, he went looking for more compute. xAI offered it. He joined in mid-2025 at the moment xAI decided to build its own image and video stack, with no existing infra, data pipeline, or model. He stayed through pre-training, post-training (reference-to-video, video extension), and a final stretch leading a small team on real-time long-horizon video generation. > *"By the time I joined, xAI was about to build video models and multimodal models. There were no infra, no data, and no model. Just a few engineers — we built it in three months and released the first model, Grok Imagine 0.9."* ## [04:40] Building Grok Imagine from Zero to One The three-month timeline surprised even Ethan. He attributes it to three factors: talent density (strong engineers who could align on a goal with minimal meetings — typically just one sync a day), xAI's existing data and inference infrastructure, and his own prior experience running the same build at NVIDIA. The bottleneck was iteration speed: how many training runs can you complete per day. With strong infra and abundant compute, bugs surface faster and each failed run costs less, so you burn through the inevitable data and pipeline errors in weeks rather than months. > *"The most important thing is talent. Everyone was very strong and clever, very close to each other toward a common goal. So that speeds up things a lot — you reduce the communication bandwidth among people."* Ethan describes a pattern where small data or pipeline bugs produce outsized quality regressions, and only fast iteration exposes them. A bug invisible at one scale becomes catastrophic at the next. The engineers who find and fix these quickly — not the ones who design the most sophisticated architecture — determine how fast a team ships. ## [11:23] How Image and Video Models Are Trained Video models require synthetic text-video pairs because internet video titles and descriptions almost never describe visual content accurately. The first step is human labeling: at NVIDIA, annotators were instructed to describe every object, character, interaction, and dialogue in a clip as exhaustively as possible. Those labels train an early VLM, which then generates captions at scale. The resulting pipeline — video to VLM to synthetic caption to (video, caption) training pair — is the foundation of both Cosmos and Grok Imagine. Image models must come first: they train faster, require less storage, and the learned representations transfer directly to video. Ethan describes building image models as building the foundation that video sits on top of. The architecture — diffusion transformer operating over VAE latents — is now standard, but the data quality and caption detail remain the primary lever for model quality. > *"Building a video model, you actually need to build an image model first. The data you need is 100% synthetic pairs of language and image, or language to video — because on the internet, videos don't naturally associate with text."* ## [20:09] Video Compression, VAEs, and Real-Time Tradeoffs Raw MP4 compression produces tokens whose latent space is incomprehensible to transformers, so the field moved to learned VAEs that create a smoother, more continuous latent space models can train on. The key design choice is how aggressively to compress the temporal dimension. Temporal compression is efficient — adjacent frames are mostly redundant — but it trades away real-time capability. Wan 2.1 uses 8x8 spatial and 4x temporal compression; generating a single token requires reconstructing four frames, making sub-200ms latency impractical. Ethan frames this as a fundamental tradeoff: high compression rates make training cheap and inference efficient for pre-rendered video, but lock out any use case that needs to respond to live user input. World models require the opposite choice. ## [23:26] Generative UI, Flipbook, and Neural OS Ethan argues that if inference were free, the logical endpoint of video generation is a complete replacement of conventional UI: instead of loading web pages from a server, a model generates them in real time in response to user intent. Flipbook, a demo that went viral, shows this literally — every element of the "browser" is generated by an image model, and clicking a link generates a new page rather than fetching one. The deeper claim is that this is not a novelty but the final form of world models applied to human-computer interaction. A traditional app is a fixed function mapping input to output; a generative UI is a model that can produce any interface the user needs without a developer having to build it first. Ethan calls this a "Neural OS," where the gap between user intent and rendered pixels closes entirely. > *"Imagine the internet doesn't exist and you type in google.com — what should a model show you? The model can imagine something. These web pages completely do not exist, so I can explore anything."* The near-term constraint is inference cost. Current video models cannot generate at interactive frame rates without significant distillation. But Ethan treats this as an engineering problem with a known solution trajectory, not a fundamental barrier. ## [33:26] The Cost of Training Large Video Models Training large video models costs roughly as much as training a medium-scale language model, but the breakdown differs. Compute is comparable, but storage and data movement dominate in ways LLM practitioners do not expect. One billion videos at 5 MB each requires five petabytes of raw storage. The VAE features that must also be stored are roughly the same size again — tens of petabytes total. On AWS S3, five petabytes runs approximately $100K per month before egress. Egress — downloading that data into the training cluster — can exceed storage costs, and each training run pulls the full dataset once. > *"Just storing the videos alone costs a lot. Five petabytes on S3 Standard is $100K per month. And egress — just to download those videos — I believe it's more expensive than storing them, and each training run you probably need to pull them once."* The implication is that video model development is gated on data infrastructure as much as on GPU hours. Teams without efficient data pipelines pay a multiplier on every experiment. ## [38:20] Distillation, GANs, and Fast Video Inference Training-time costs are largely fixed; the inference-time story is more tractable. Step distillation — training a small model to replicate the outputs of a large teacher in far fewer denoising steps — cuts inference cost by 10-25x. Flow-matching models trained to convergence need around 100 steps; production models typically run in 4-8. At the extreme, simple image-to-image tasks can run in a single step. The intuition Ethan offers: the teacher model must learn the full distribution of internet video, which is arbitrarily complex. The distilled student only needs to match the teacher, which is a fixed and much simpler target. Consistency models and LCM-style approaches follow the same logic. In Cosmos, production serving used 4-step and 8-step variants depending on quality requirements. GANs remain relevant as discriminators: a GAN discriminator can enforce photorealism constraints during distillation that pure score-matching loss misses, and Ethan notes that consistency models and GANs are converging on similar practical deployments even if their theoretical motivations differ. ## [42:37] Audio-Video Generation and Grok Imagine 0.9 Grok Imagine 0.9 was the first audio-video joint generation model deployed at scale. The core difficulty is modality alignment: text-video pairs are relatively abundant; text-audio pairs are rare; audio-video pairs aligned at the semantic level are almost nonexistent at scale. Speech tokens are quasi-discrete and can be modeled with language-like approaches, but music is continuous and requires a completely different representation. Training the joint model required building synthetic audio caption pipelines from scratch, with human annotation where VLMs failed — which was often, especially for music. Aligning all three modalities — text, video, and audio — without either degrading video quality or audio realism is what Ethan calls the hardest part of the project. > *"Audio has two components: a discrete component — language — and a continuous component — music. The music is completely different; you cannot model it with discrete tokens. That's the hard part, not to mention we have to align text, video, and audio together."* ## [49:50] What Makes a World Model? Ethan's definition has three components: real-time, interactive, and long-horizon video generation. He treats these as independent requirements, each of which most current models fail. Real-time means generating at display frame rates — 60fps for casual use, 300fps for gaming, 200ms response latency for digital humans. Current video models cannot do this; the VAE's temporal compression alone introduces latency that makes sub-200ms responses nearly impossible without architectural changes. Interactive means the model can accept any input modality the user can provide — keyboard, mouse, voice — and respond coherently. Long-horizon means maintaining consistent physical laws, character identity, and causal logic across minutes, not seconds. > *"World model is real-time, interactive, long-horizon video. Current video models can do none of these three things fully. That's why they're not world models yet."* ## [57:07] Reference Videos, Long Context, and Video Memory The parallel to language model context scaling is direct: video models are in the 2,000-8,000 token era, and will need to scale to million-token-equivalent contexts to generate coherent long videos. Ethan describes the reference-to-video feature he built at xAI (analogous to Cameo) as a mechanism for injecting selected history into the model's context rather than carrying the full video forward. FramePack's heuristic — storing the last second of video at full resolution while compressing earlier frames progressively — points toward the right direction: the model selects relevant context from its history rather than brute-forcing the full sequence. Ethan expects this context management to become part of the model itself rather than remaining a harness-level heuristic, the same way KV cache management is disappearing into model internals. ## [61:27] xAI Culture, Research, and First-Principles Building swyx notes that xAI communicates its research poorly relative to what the work actually demonstrates — the blog post accompanying Grok Imagine describes high-level capabilities without the technical depth Ethan has just spent an hour covering. Ethan is diplomatic but agrees that different labs have different communication styles. The xAI working culture he describes is minimalist: few meetings, no bureaucratic overhead, direct access to leadership judgment on technical decisions, and extreme iteration speed enabled by a strong infra team. The tradeoff is that company priorities shift fast, which is part of what eventually pushed him toward independent research. First-principles thinking — starting from the physics of the problem rather than from what competitors have shipped — runs through the team's approach to both model architecture and product. > *"Everything you just described is state-of-the-art. Like no one else has done it. And then you just put this blog post with the cookies. I'm like, this is not enough."* ## [71:01] AI Safety, Watermarking, and Prompt Rewriting Grok Imagine deployed watermarks in all jurisdictions requiring them and built takedown pipelines integrated with xAI's social platform infrastructure. On watermarking technology, Ethan is skeptical of SynthID's long-term robustness: the technique is documented publicly, and users on Reddit have already reverse-engineered the exact frequency pattern Google applies and can strip it from any generated image. He expects watermark detection to become an arms race. On prompt rewriting: video diffusion models take instructions literally. If a user types "a cat," the model generates a stationary cat on a white background with no motion, because the training data pairs were maximally detailed descriptions of physical scenes. Production systems layer a large language model as a prompt upsampler — converting sparse user instructions into the detailed physical descriptions the video model was trained on. This is one of the reasons Ethan argues language models are increasingly central to video quality. ## [74:26] Video Agents and AI-Assisted Creation Ethan's central claim from the hook: visual intelligence now mostly comes from language. The diffusion model architecture has largely converged; the gains come from larger, smarter LLMs that rewrite prompts, plan video sequences, call editing tools, and stitch clips together. In Cosmos, the prompt rewriter was larger than the video model itself. Video agents extend this: instead of generating a complete video in one shot, an agent plans the production, calls video generation models as tools alongside deterministic editing operations (text overlays, color grading, cuts), and iterates until the output meets a specification. Ethan predicts that by end of 2025, video agent output will reach production-grade quality — presentable video generated without a human editor in the loop. > *"The visual intelligence are actually mostly coming from language. Every time you see improvement on these models, I would say mostly the gain comes from language model, not coming from the video model itself."* ## [88:48] Why Language Models Unlock Better Video LLMs prompt video models better than humans do, because AI models understand AI models' training distributions. A language model knows that a diffusion model needs explicit physical descriptions, not poetic shorthand — and can generate the right prompt format automatically. Beyond prompting, agents can use deterministic video editing tools for precision operations (exact text overlays, frame-accurate cuts) that probabilistic diffusion models handle poorly, keeping the stochastic model focused on generation and delegating precision to tools. Ethan's timeline: video agent output at production quality by end of 2025, with the inflection point visible in work already shipping. ## [92:31] Robotics, Physical AI, and Embodied World Models Ethan's robotics prediction inverts the usual framing: physical AI may be solved not by deploying robots in the real world but by video world models becoming so capable at simulating physical environments that they effectively provide embodied experience. Once a model can control computer interfaces in real time with full causal understanding, extending that to robotic control becomes a matter of adding one more tool. The path from screen-interacting video model to robot controller may be shorter than the path from current robot learning systems to the same capability. ## [93:54] Why Ethan Left xAI Research ambitions and company priorities diverged. xAI's focus shifted in ways that made certain research directions — particularly on the language model side — impractical from inside. Ethan also notes that the insight driving his departure is the same one underlying his "big claim": if language models are now the primary driver of video quality, the most impactful work to do is on language models, not video models. He frames leaving not as dissatisfaction but as following the evidence about where the leverage is. ## [95:32] Self-Managed Context and the Future of LLMs Ethan's active research question: language models that are aware of their own context state and manage it autonomously, rather than relying on harness-level heuristics like automatic compaction at 80% fill. He draws the parallel to video models struggling with long-horizon generation — the same context management problem appears in both modalities. He points to Claude Code's practice of appending the current timestamp to user messages as an early example of making models context-aware, and expects this pattern to be absorbed into model training rather than remaining an external scaffold. > *"The language models are not aware of how long their own context length is. Once they hit like 80% or something, automatic context compaction is getting triggered, and the model is not aware of that when it's working."* ## [99:59] Ethan's Career Path and Closing Thoughts Ethan traces a decade of transitions: ResNet-era image recognition with the original authors at NVIDIA, self-supervised learning at Facebook AI Research, scaling at NVIDIA Cosmos, extreme-scale compute at xAI. He was rejected from every top PhD program despite first-author papers at top conferences, which pushed him into industry. In hindsight he reads his career as consistently following the scaling frontier — from image recognition to SSL to video to LLMs — and argues that within ML, domain switching is far more tractable than practitioners believe. > *"Within ML, it's actually easier to switch than you think. A lot of people have manifested that 'I work on computer vision, I always have to work on computer vision.' But from my experience, the fundamentals transfer."* ## Entities - **Ethan He** (Person): Former xAI researcher who built Grok Imagine from zero; previously led NVIDIA Cosmos world model; now focused on LLM research - **swyx** (Person): Latent Space co-host; conducts technical interviews on AI engineering and research - **Vibhu Viswanathan** (Person): Latent Space co-host; co-interviewer for this episode - **Grok Imagine** (Software): xAI's image and video generation product; first model (0.9) was the first large-scale audio-video joint generation system - **NVIDIA Cosmos** (Software): Open-source video foundation model for robotics simulation; Ethan's project before xAI; released end of 2024 - **xAI** (Organization): Elon Musk's AI lab; known for fast iteration culture and extreme compute resources - **Flipbook** (Software): Viral demo of real-time generative UI; all interface elements generated by image model in real time - **SynthID** (Software): Google's AI watermarking technology; Ethan notes its pattern has been publicly reverse-engineered - **Step distillation** (Concept): Technique to train a model to replicate a teacher's output in far fewer denoising steps; reduces inference cost 10-25x - **VAE** (Concept): Learned video compression creating smooth latent spaces; temporal compression is efficient but creates real-time latency tradeoffs - **World model** (Concept): Ethan's definition — real-time, interactive, long-horizon video generation; distinct from standard video generation - **Video agents** (Concept): Systems where LLMs orchestrate video generation models, editing tools, and deterministic operations to produce production-quality video - **FramePack** (Concept): Progressive temporal compression approach for long-context video generation; stores recent frames at full resolution, compresses older history

Alex Rives — directeur scientifique de BioHub et chercheur à l'origine des modèles ESM-1 à ESM-3 chez Meta FAIR — rejoint Brandon et RJ Honicky pour expliquer pourquoi il parie depuis huit ans que le passage à l'échelle d'un modèle de langage masqué sur des séquences de protéines permettrait d'en comprendre la structure, la fonction et la conception. L'épisode couvre le passage des données UniRef à la métagénomique qui a restauré la loi d'échelle d'ESMC, l'atlas de caractéristiques par autoencodeurs épars qui reproduit un siècle de taxonomie biochimique sans y avoir été entraîné, et le premier succès rapporté dans la conception d'anticorps à chaîne unique à visée thérapeutique par recherche dans un world-model. Rives présente également l'initiative Virtual Biology à 500 millions de dollars de BioHub et les principes qui, selon lui, mèneront à des modèles généralistes de la cellule. ## [00:00] ESMC conçoit des anticorps — un aperçu Ce clip d'ouverture est tiré d'un moment ultérieur de l'entretien, où Rives décrit en pleine phrase l'approche d'ESMC pour une biologie programmable. Il évoque la recherche dans un world-model de protéines pour satisfaire des critères de conception, et mentionne que l'équipe a conçu des mini-binders et, surtout, des fragments d'anticorps à chaîne unique (SCFVs) avec des affinités de liaison d'intérêt thérapeutique. Ce clip précède l'introduction formelle — un signal de ce vers quoi l'épisode se dirige. ## [00:33] La Bitter Lesson arrive pour les protéines Brandon et RJ présentent Alex comme peut-être "la personne la plus convaincue par la bitter lesson dans la biologie des protéines en ce moment." Rives accepte l'étiquette. Il retrace sa conviction jusqu'en 2018, quand son équipe chez Meta FAIR a entraîné le premier transformer de langage sur des séquences de protéines par prédiction de tokens masqués, et vu émerger des représentations structurelles et fonctionnelles sans aucune supervision explicite. L'intuition clé, empruntée au papier de 1954 de Zellig Harris sur la structure distributionnelle, est que les contextes dans lesquels un acide aminé peut apparaître sont déterminés par la structure, la fonction et le rôle évolutif de la protéine. Cette pression statistique, appliquée à des milliards de séquences issues de l'ensemble du vivant, devrait forcer un modèle à apprendre les variables cachées qui gouvernent la biologie des protéines. > *"Je crois aux lois d'échelle."* ## [06:00] La lignée ESM : d'ESM2 à ESMC Rives retrace quatre générations d'ESM. ESM2 montrait des gains à l'échelle, mais atteignait des rendements décroissants à 10 milliards de paramètres — non pas parce que le modèle était saturé, mais parce que les données l'étaient. UniRef, la base de données de référence pour les protéines, capture des organismes cultivés en laboratoire et surreprésente la biologie humaine. La solution pour ESMC a été la métagénomique : des séquences extraites de sources hydrothermales, de sols polaires et d'eaux usées, assemblées à partir de lectures d'ADN environnemental brut sans assignation à un organisme, contigs partiels inclus. L'ajout de milliards de séquences métagénomiques à l'entraînement a restauré une loi d'échelle log-linéaire propre, les runs à plus petite échelle prédisant avec précision la fidélité représentationnelle du modèle phare à 6 milliards de paramètres. > *"Les rendements décroissants à l'échelle n'existent plus. ESM2 était limité par les données, pas par le calcul."* ESMC est essentiellement un transformer standard avec des objectifs de masquage classiques — sans MSA à la AlphaFold, sans biais inductifs géométriques. Brandon et Rives débattent brièvement de l'utilité de l'architecture multi-piste d'ESM3 ; Rives dit que les deux paradigmes ont leur place, mais que les résultats d'ESMC suggèrent que ces a priori n'étaient pas déterminants à cette échelle de données. ## [18:30] Interprétabilité mécanistique et l'atlas des caractéristiques protéiques À l'aide d'autoencodeurs épars entraînés sur toutes les couches de la famille de modèles ESMC (300M, 600M, 6B), l'équipe de BioHub a extrait la géométrie intrinsèque des caractéristiques de l'espace de représentation des protéines. Ce qui émerge correspond étroitement à la hiérarchie réductive que la biologie a développée expérimentalement sur un siècle — de la chimie de base des acides aminés jusqu'aux motifs structuraux, aux familles de domaines et aux grandes thématiques fonctionnelles — sans qu'aucune de ces taxonomies n'ait été intégrée à l'entraînement. > *"Le choix d'un acide aminé est en quelque sorte complètement entrelacé avec le choix de tous les autres acides aminés de la séquence. Pour bien faire cela, le modèle commence à avoir ces variables cachées qui représentent la biologie."* Un résultat concret : le modèle encode le coude nucléophile — un motif catalytique supposé avoir évolué indépendamment dans plusieurs familles de protéines non apparentées — comme une seule caractéristique qui s'active dans chacune d'elles. L'équipe a également construit un atlas structurel de 6,8 milliards de protéines non redondantes avec des structures prédites pour 1,1 milliard de représentants de clusters, et utilisé les caractéristiques SAE pour relier des systèmes d'édition génique évolutivement distants. Certaines protéines intégrées à ces clusters n'ont aucune fonction connue ; Rives les traite comme une file d'attente de découvertes. La première version de l'atlas ESM a déjà été utilisée par un groupe externe pour trouver un nouveau système d'édition génique. ## [35:30] Concevoir des anticorps avec ESMC Rives décrit la conception de protéines comme une recherche dans un world-model : inverser le modèle génératif pour trouver des séquences satisfaisant des critères de liaison ciblés. Les mini-binders sont désormais courants ; les nanobodies et les SCFVs restent plus difficiles pour les méthodes basées sur la prédiction de structure, car l'évolution des anticorps maximise la diversité plutôt que de converger vers un repliement contraint, ce qui rend les approches fondées sur les MSA moins utiles. ESMC, entraîné sur cette diversité à grande échelle, est précisément là où la représentation devrait être la plus riche. > *"Les anticorps ne vont probablement pas bénéficier de l'information évolutive de la même façon que la prédiction de la topologie structurale d'une molécule."* L'équipe rapporte que des designs de SCFVs atteignent une affinité de qualité thérapeutique en un petit nombre d'essais, et note que les SCFVs peuvent être reformatés en IgG complets. ESMFold 2 — la tête de prédiction de structure construite sur les représentations ESMC — s'exécute en quelques secondes par séquence sans MSA, rendant faisable la cartographie de multimères à l'échelle du protéome entier. Rives affirme que le modèle est actuellement à l'état de l'art pour la prédiction de multimères en open-weight. ## [42:00] La vision de BioHub : vers une biologie programmable Six mois après avoir rejoint BioHub, Rives articule la structure de l'institution : une organisation philanthropique construisant simultanément la biologie expérimentale de pointe, les technologies de mesure de pointe et l'IA de pointe, sous un mandat de science ouverte. Il situe l'objectif final dans des modèles prédictifs personnalisés de la physiologie — non pas un médicament, mais un système capable de retracer les événements moléculaires au niveau des protéines jusqu'aux circuits cellulaires et à la manifestation de la maladie dans un génome humain spécifique. > *"Nous construisons une institution scientifique pour ce nouveau paradigme."* Il cartographie les niveaux de complexité biologique qui doivent être modélisés successivement : les protéines (génération actuelle), la cellule (prochaine étape), les tissus et systèmes, la physiologie. Passer des protéines aux cellules nécessite des données qui n'existent pas encore et des approches de modélisation qui n'ont probablement pas encore été inventées. Les modèles actuels de "cellule virtuelle" généralisent mal — ils représentent bien les données d'entraînement, mais échouent à prédire les résultats dans des contextes interventionnels nouveaux et non observés. > *"Ils ont une capacité très limitée à prédire ce qui se passera lorsqu'on effectue une intervention nouvelle dans un contexte nouveau et non observé."* ## [57:00] L'initiative Virtual Biology et la mise à l'échelle des données cellulaires BioHub a récemment annoncé 400 millions de dollars pour la génération de données interne et les technologies de mesure, plus 100 millions de dollars pour catalyser des efforts externes — c'est l'initiative Virtual Biology. Rives la présente comme un amorçage : le volume de données réellement nécessaire est bien plus important, et l'espoir est que l'engagement de BioHub déclenche un investissement plus large de la communauté scientifique. Il identifie trois principes pour les données : la vitesse (les données sur les protéines ont pris un demi-siècle ; la cellule ne peut pas attendre aussi longtemps), la généralisation (la distribution d'entraînement doit couvrir une vaste diversité d'interventions sur des types cellulaires et des contextes variés, analogue à l'étendue métagénomique pour les protéines), et le retour d'information (des boucles expérimentales actives guidées par les prédictions du modèle — quelque chose comme le RLVR appliqué à la biologie en laboratoire humide). Le séquençage par perturbation, la transcriptomique spatiale et la mesure unicellulaire multimodale sont les technologies évolutives prêtes à fonctionner dès maintenant. Sur le calcul : ESMC a été entraîné sur environ un milliard de séquences. On estime qu'il en existe environ 100 milliards, et le modèle n'a pas encore pleinement exploité même les 6,8 milliards de l'atlas actuel. Une multiplication par 100 de la puissance de calcul aiderait, mais uniquement associée à une mise à l'échelle proportionnelle des données. Rives laisse ouverte la question de savoir quand les rendements décroissants apparaîtront empiriquement — la courbe d'ESM2 semblait saturée jusqu'à ce que les données métagénomiques l'effacent. > *"Nous devons trouver comment faire cela en quelques années. La vitesse à laquelle l'IA généraliste se développe signifie que la biologie sera fondamentalement limitée par la science expérimentale et les données."* ## Entités - **Alex Rives** (Personne) : Directeur scientifique de BioHub ; architecte de ESM-1, ESM-2, ESM-3, ESMC et ESMFold 2 ; anciennement Meta FAIR. - **Brandon** (Personne) : Co-animateur de la sous-série Latent Space AI for Science ; affilié à Atomic AI (ARN thérapeutiques). - **RJ Honicky** (Personne) : Co-animateur ; CTO et fondateur de Miro Omix. - **ESMC** (Logiciel) : Modèle de langage de protéines de quatrième génération de BioHub/EvoScale ; 300M–6B paramètres ; entraîné sur ~1B séquences dont des données métagénomiques ; open source sous licence MIT. - **ESMFold 2** (Logiciel) : Modèle de prédiction de structure construit sur les représentations ESMC ; sans MSA, inférence en quelques secondes par séquence ; meilleure prédiction de multimères en open-weight. - **ESM** (Logiciel) : Evolutionary Scale Modeling — la lignée multi-générationnelle de modèles de langage de protéines (ESM-1, ESM-2, ESM-3, ESMC) développée par l'équipe de Rives. - **Autoencodeurs épars / SAE** (Concept) : Outil d'interprétabilité mécanistique utilisé pour extraire la géométrie intrinsèque des caractéristiques de l'espace de représentation d'ESMC ; révèle des hiérarchies biologiquement interprétables sans supervision. - **Bitter Lesson** (Concept) : L'argument de Richard Sutton selon lequel les méthodes générales exploitant le calcul et les données surpassent systématiquement les méthodes encodant des connaissances du domaine ; appliqué ici à la mise à l'échelle en biologie des protéines. - **Séquençage métagénomique** (Concept) : Séquençage d'ADN environnemental capturant la diversité microbienne et virale sans culture ; l'expansion de données qui a restauré la loi d'échelle d'ESMC là où UniRef avait atteint sa limite. - **BioHub** (Organisation) : Chan Zuckerberg BioHub ; organisation philanthropique construisant des outils de science ouverte à l'intersection de la biologie expérimentale, des technologies de mesure et de l'IA. - **Virtual Biology Initiative** (Concept) : L'engagement de 500 millions de dollars de BioHub (400M en interne, 100M en externe) pour générer les données à l'échelle cellulaire nécessaires à l'entraînement de modèles généralistes de la cellule. - **AlphaFold** (Logiciel) : Système de prédiction de structure de DeepMind ; utilise les MSA et des biais inductifs géométriques ; mis en contraste avec l'approche sans MSA d'ESMC. - **UniRef** (Logiciel/Base de données) : Base de données curatée de référence pour les séquences de protéines ; données d'entraînement d'ESM2, identifiée ultérieurement comme le goulet d'étranglement à l'origine du plateau de mise à l'échelle d'ESM2. - **Coude nucléophile** (Concept) : Motif structural catalytique présent dans plusieurs familles de protéines évolutivement non apparentées ; encodé comme une seule caractéristique ESMC s'activant dans chacune d'elles. - **Zellig Harris** (Personne) : Linguiste ; son article de 1954 "Distributional Structure" a formulé l'idée que les contextes de mots encodent le sens — un précurseur théorique que Rives cite pour expliquer pourquoi les statistiques de contexte des acides aminés devraient encoder la fonction biologique.

Dans cet épisode éclair, swyx s'entretient avec Sunil Pai — responsable de la plateforme développeur chez Cloudflare et, selon swyx, créateur de Code Mode — autour de trois fils conducteurs : le pari infrastructurel de Cloudflare sur les Durable Objects et les Dynamic Workers comme socle pour les agents IA, le malentendu Twitter avec Vercel qui a failli convaincre Sunil que sa carrière était terminée, et pourquoi forker du code relève du respect plutôt que de l'agression. Sunil conclut sur un défi direct : arrêtez de construire des frameworks d'agents incrémentaux et construisez de la science-fiction. ## [00:00] Qui a inventé Code Mode ? La vidéo s'ouvre sur un générique de trois secondes. Ce qui suit immédiatement — swyx présentant Sunil comme le « créateur de Code Mode », Sunil acceptant le crédit avec une grandeur faussement solennelle, affirmant y penser depuis l'enfance — est l'échange d'ouverture que ce chapitre couvre. Pure complicité entre deux vieux amis, pas une accroche tirée de plus loin dans l'épisode. ## [00:03] Introduction et parcours de Sunil Pai swyx réintroduit Sunil comme un vieil ami et conférencier principal à l'AIE Europe. Ce bref rattrapage pose le cadre de ce qui suit : l'attention de Sunil se porte sur la plateforme d'agents IA de Cloudflare, et le lancement récent d'Anthropic Cloud Managed Agents lui offre un point de comparaison concret à réfuter. > *"Je voulais juste faire le point sur tout ce qui se passe du côté de Cloudflare."* ## [00:30] Les nouveaux agents cloud managés en discussion Le produit Anthropic Cloud Managed Agents — une plateforme pour construire et déployer des agents de longue durée — sert de point de départ à Sunil. Il dit apprécier l'équipe Anthropic et trouver le produit intéressant, mais sa réaction à la lecture des spécifications était compétitive : Cloudflare peut faire mieux. swyx lui demande ce que Cloudflare a concrètement pour rendre cette affirmation crédible. > *"J'ai regardé le produit et je me suis dit que je voulais concurrencer. Je pense qu'on peut faire quelque chose de mieux avec Workers et Durable Objects."* ## [01:10] L'infrastructure de Cloudflare : Durable Objects et Dynamic Workers Sunil identifie deux primitives qu'il juge indispensables à toute plateforme d'agents. Les Durable Objects sont des unités serverless avec état — selon lui, la première implémentation au niveau infrastructure du modèle acteur, plutôt qu'une bibliothèque en espace utilisateur. Les Dynamic Workers sont la réponse de Cloudflare pour exécuter du code généré par LLM en toute sécurité : un `eval` repensé, sans temps de démarrage, avec une surface API configurable et le trafic sortant verrouillé par défaut. Ensemble, ils permettent d'exécuter les étapes d'un agent dans un environnement de calcul isolé, sans lancer de VM complète. > *"C'est la première implémentation au monde du modèle acteur au niveau infrastructure, pas en espace utilisateur."* ## [02:34] L'approche de Cloudflare pour l'architecture des agents IA Le serveur MCP de Cloudflare, développé par le collègue Matt Carey, illustre les Dynamic Workers en pratique. L'API Cloudflare compte 2 600 endpoints — exposer un outil par endpoint détruirait n'importe quelle fenêtre de contexte LLM. À la place, le serveur compresse tout en deux appels d'outils : `search` et `execute`, tous deux appuyés sur du JavaScript s'exécutant dans un isolate. L'agent soumet du code, l'isolate l'exécute, le résultat revient — sans aller-retour, avec vérification de types. > *"En un seul appel d'outil, sans aller-retour avec le LLM, et c'est vérifié en types — les LLMs s'avèrent excellents pour exécuter du code."* ## [03:40] L'avenir des logiciels agentiques et la standardisation du "harness" swyx demande si le concept de harness tiré des spécifications Anthropic pourrait devenir un standard multiplateforme. La réponse de Sunil : personne n'a encore construit le React des agents IA. Il tire délibérément l'analogie React 2013 — les gens ont quitté la conférence JSConf en accusant Facebook de haïr JavaScript, et pourtant React a défini tous les frameworks UI qui ont suivi. Aujourd'hui, chacun construit son propre harness à sa façon, et rien n'est reproductible entre les langages, les entreprises et les infrastructures. swyx avance que les skills — de simples fichiers markdown — pourraient déjà être cette couche unificatrice ; Sunil trouve l'idée genuinement séduisante mais s'inquiète du plafond de spécificité. > *"C'est tellement difficile, mais la façon dont je le formule dans ma tête, c'est que personne n'a encore construit le React."* ## [06:11] Le phénomène des "slop forks" et la culture open-source swyx soulève les « slop forks » — des forks de projets populaires générés par IA — et Sunil s'anime. Dans sa vision, forker est un geste de prestige et de respect, pas un vol. L'écosystème React a grandi grâce aux forks. Il encourage quiconque voudrait construire quelque chose en compétition avec le Cloudflare Agents SDK à se lancer : tout le monde y gagne. > *"Forker est un grand signe de prestige et de respect dans ma culture."* ## [06:36] Le malentendu Vercel / Cloudflare sur les réseaux sociaux À JSConf España, Sunil a rencontré Harvey de Vercel et a apprécié passer du temps avec lui. Il a découvert Just Bash — une implémentation purement JavaScript de Bash créée par Vercel Labs — et voulait la porter sur Cloudflare. Il a pointé Opus sur le code pendant le déjeuner, récupéré 5 000 lignes de code, et prévu de tout nettoyer avant d'envoyer une vraie PR le lundi. Il s'est endormi, s'est réveillé avec des messages de sa direction Cloudflare lui demandant s'il avait vu Twitter : le CTO de Vercel avait publiquement critiqué le travail, présentant cela comme une démarche corporate plutôt qu'un projet personnel. Sunil a répondu simplement, expliqué le contexte, puis regardé la moitié d'internet se précipiter pour le défendre. > *"Je vais sur Twitter et le CTO de Vercel descend mon travail en disant que 'c'est Cloudflare qui a fait ça'."* ## [09:45] L'importance du fork dans le développement logiciel swyx relie l'incident Vercel à un schéma plus large : une base de code diffusée que quelqu'un a réécrite en Python pour échapper à la licence (les juristes ont quand même conclu à une œuvre dérivée). L'argument central de swyx est que les slop forks méritent d'être encouragés — forker une dépendance, la vendre en interne, la posséder — pour éviter la rupture soudaine causée par un upstream comme les problèmes LiteLLM ou Axios. Sunil acquiesce : avant NPM, les logiciels se répandaient sur Usenet exactement selon ce schéma, et raccourcir le cycle de fork, c'est simplement continuer cette tradition. > *"Le fork est fondamental à la façon dont nous construisons les logiciels."* ## [12:04] La nature adversariale des dépôts open-source modernes Le Cloudflare Agents SDK a dû fermer entièrement les contributions par pull request ; seules les issues sont désormais acceptées. Sunil parle aux mainteneurs open-source lors de la conférence, qui décrivent la même situation : les dépôts sont devenus un territoire adversarial, et le vecteur d'attaque le plus redoutable est le faux rapport de sécurité qui paraît entièrement légitime jusqu'à une lecture attentive. swyx relie cela à une conférence du matin par Peter de Claude Code — la principale surface d'attaque actuelle est une dépendance compromise qui s'introduirait dans Claude Code, donnant accès à tous les développeurs qui l'utilisent. > *"Les dépôts open-source sont devenus si adversariaux que les gens ont presque peur de gagner en popularité dans cet espace."* ## [13:04] Réflexions finales et invitation à l'originalité La conclusion de Sunil est directe : arrêtez de construire le dixième framework d'agents. Construisez de la science-fiction. Construisez quelque chose pour votre famille. Utilisez l'Agent SDK, mais pour quelque chose où l'infrastructure et les LLMs vous mettent presque en échec — c'est là que vit le prochain bond en avant. swyx clôt avec un retour à la formule « alpha thought leading » que Sunil avait inventée à React Rally 2018. > *"Construisez de la SF. Construisez pour votre famille. Vous avez tellement de pouvoir pour changer le monde — soyez originaux."* ## Entités - **swyx** (Personne) : Animateur de Latent Space ; vieil ami de Sunil ; a inventé « alpha thought leading » après une boutade de Sunil à React Rally 2018. - **Sunil Pai** (Personne) : Responsable de la plateforme développeur chez Cloudflare ; crédité par swyx comme créateur de Code Mode ; conférencier principal à l'AIE Europe. - **Cloudflare** (Organisation) : Société de plateforme cloud ; construit l'infrastructure pour agents sur les Durable Objects et les Dynamic Workers. - **Anthropic** (Organisation) : Société d'IA ; a lancé Cloud Managed Agents, le produit face auquel Sunil positionne Cloudflare. - **Vercel** (Organisation) : Société de cloud frontend ; Sunil utilise leur AI SDK ; à l'origine du malentendu Twitter. - **Durable Objects** (Logiciel) : Primitive serverless avec état de Cloudflare ; Sunil affirme qu'il s'agit de la première implémentation au monde du modèle acteur au niveau infrastructure. - **Dynamic Workers** (Logiciel) : Fonctionnalité Cloudflare pour exécuter du JavaScript généré par LLM ou par l'utilisateur dans un isolate sécurisé, sans démarrage à froid. - **Just Bash** (Logiciel) : Projet Vercel Labs — une implémentation purement JavaScript de Bash — que Sunil était en train de porter sur Cloudflare lors de l'incident Twitter. - **MCP** (Concept) : Model Context Protocol ; le serveur MCP de Cloudflare compresse 2 600 endpoints API en deux appels d'outils grâce aux Dynamic Workers. - **Slop forks** (Concept) : Forks de projets existants générés par IA ; Sunil les cadre comme la continuation de la culture du fork open-source — un signe de respect, pas du plagiat.

Enregistré en direct à AI Engineer London, swyx s'entretient avec Omar Sanseviero — responsable Developer Experience chez Google DeepMind — pour un sprint de 30 minutes sur les nouveautés architecturales de Gemma 4, la stratégie open model de Google, et les prochaines zones de développement de l'équipe DevEx. Omar lève le voile sur les embeddings par couche, explique pourquoi l'engouement pour le fine-tuning s'est tari, ce que l'arrivée de Kaggle dans DeepMind change concrètement pour les benchmarks, et si l'« auto-recherche » est une réalité ou du pur marketing. ## [00:00] Introduction à Gemma 4 et périmètre de l'équipe Le pitch d'Omar en une phrase : Gemma 4 est « le modèle open le plus capable que nous ayons publié à ce jour », avec une contrainte forte — maximiser l'intelligence par paramètre et assurer un support multimodal complet, tout en maintenant un poids total compatible avec l'inférence locale. > *"Nous avons vraiment cherché à compresser un maximum d'intelligence par paramètre."* ## [00:23] Paramètres effectifs vs. paramètres actifs Le grand choix architectural des petits modèles Gemma 4 : une table d'embeddings par couche insérée dans chaque bloc transformer. Comme il s'agit d'une table de correspondance et non d'une multiplication matricielle, les 3 milliards de paramètres d'embedding n'ont jamais besoin d'être chargés en mémoire GPU — ils restent sur CPU ou sur disque, tandis que seuls les 2 milliards de paramètres actifs font le calcul à chaud. Omar reconnaît franchement que cette astuce est pensée pour le on-device : à plus grande échelle, on préférera des architectures denses ou MoE. > *"Le modèle Gemma 4 est E2B. Cela signifie qu'il charge effectivement 2 milliards de paramètres dans le GPU. Il en possède en réalité près de 5 milliards, mais ces 3 milliards peuvent résider sur le CPU ou sur le disque."* ## [01:43] Cas d'usage on-device et intégration de Gemini Nano Les téléphones Pixel et les Samsung haut de gamme embarquent Gemini Nano dès la sortie de boîte — Gemini Nano étant entraîné sur Gemma 3N, l'architecture conçue spécifiquement pour les contraintes des smartphones. Le même mécanisme de déchargement de paramètres s'applique à ces variantes plus légères. Quand swyx demande si cela s'étend au palier 29B–31B, Omar répond sobrement : « nous menons beaucoup d'expériences — restez à l'écoute. » > *"Quand vous achetez ces téléphones haut de gamme, vous pouvez déjà utiliser Gemini directement."* ## [03:14] Dans les coulisses d'un lancement de modèle et de l'écosystème développeur L'équipe Gemma est plus petite que ce qu'on imagine — deux ou trois chefs de produit, un chargé de communication, des ingénieurs et des chercheurs. Ce qui rend un lancement complexe, c'est le réseau externe : 50 partenaires (llama.cpp, Ollama, MLX, Hugging Face, vLLM, Nvidia, AMD et d'autres) coordonnés en parallèle, plus la collaboration interne avec Google Cloud, Vertex, ADK et Android. Le lancement de Gemma 4 comprenait aussi une intégration native au mode agent d'Android Studio, permettant aux développeurs de faire tourner Gemma 4 en inférence hors ligne pour l'assistance au code. > *"Nous avons eu près de 50 partenaires externes pour le lancement de Gemma 4, ce qui en a fait le lancement le plus complexe."* ## [04:29] Usage hors ligne vs. API et croissance future des modèles La distinction vie privée / hors ligne existe, mais elle n'est pas toute l'histoire. Omar trace une ligne plus nette : les modèles locaux sont aujourd'hui excellents en capacités (function calling, suivi d'instructions, tâches agentiques), mais restent inférieurs en densité de connaissances — pour un rappel fiable de faits rares, il faut encore un grand modèle. Son pari sur un à deux ans : un modèle de niveau Gemini Pro tournant entièrement on-device, rendant possibles des expériences aujourd'hui réservées à une connexion API. > *"Je pense que dans un à deux ans, vous pourrez faire tourner un modèle puissant de niveau Gemini Pro directement sur votre téléphone."* ## [06:26] Capacités multimodales de Gemma 4 et ses limites Gemma 4 hérite de la pile de recherche de Gemini 3, ce qui donne même au modèle 2B une compréhension audio (reconnaissance vocale, traduction speech-to-text, questions-réponses sur des clips audio) et visuelle (détection d'objets, pointage, sous-titrage). Deux lacunes qu'Omar cite explicitement : la segmentation d'image est absente, et passer simultanément vidéo et audio dans un même prompt n'est pas encore supporté — les deux flux doivent entrer séparément. La génération vocale native est explorée, mais rien n'a été annoncé. > *"Nous pouvons comprendre une entrée vidéo ou une entrée audio séparément, mais si vous voulez passer dans le même prompt une partie visuelle et une partie audio, il nous reste encore des améliorations à apporter."* ## [08:08] Insights sur le tokenizer multilingue Le tokenizer de Gemma est le même que celui de Gemini — un choix qui lui confère une base multilingue exceptionnellement solide sur 140 langues. Résultat concret relevé par Omar : partez de Gemma 3 comme modèle de base, affinez-le sur une langue d'Asie du Sud-Est comme le vietnamien, et il surpasse des modèles de base qui scorent pourtant mieux sur les benchmarks anglais. Le tokenizer capte des tokens adaptés à chaque langue plutôt que de forcer les écritures non latines à travers des fragments de sous-mots calibrés pour l'anglais. > *"Si vous affinez l'ensemble de ces modèles sur une langue spécifique d'Asie du Sud-Est — le vietnamien, par exemple — Gemma donnera de meilleurs résultats, même si les autres modèles de base étaient potentiellement plus performants."* ## [09:30] L'équipe Developer Experience de Google à AI Engineer Londres est la ville natale de DeepMind, et y débarquer avec une équipe complète pour AI Engineer Europe était un acte délibéré. Omar a emmené des chercheurs travaillant sur Gemma 4, la génération de texte par diffusion, la robotique, le ML on-device et Android — pas une simple tournée DevEx. swyx résume le périmètre sans détour : « C'est le labo avec le plus grand spectre. Vous faites tout, même des dauphins. » > *"Nous avons amené des personnes de la robotique, de la recherche, d'Android. C'est vraiment stimulant de montrer tout ce que l'entreprise construit."* ## [10:42] Introduction aux domaines de recherche : modèles de diffusion pour le texte Google a annoncé Gemini Diffusion à Google I/O — un transformer de diffusion qui génère du texte (et non des images) à une vitesse nettement supérieure au décodage autorégressif. L'avis honnête d'Omar : la qualité reste inférieure aux baselines autorégressive, et affiner des transformers de diffusion est plus difficile parce que les changements de distribution affectent le routage autrement. swyx esquisse une architecture plausible où les modèles de diffusion joueraient le rôle d'exécuteurs système-1 rapides pendant que les modèles autorégressifs gèrent la planification complexe — Omar juge l'idée plausible, mais prématurée. > *"Pour l'instant, c'est encore très expérimental. La qualité du modèle est encore un peu en deçà de ce qu'on obtiendrait avec un modèle autorégressif classique."* ## [13:37] État actuel du fine-tuning et tendances communautaires Les communautés de fine-tuning avaient atteint leur sommet vers 2023 ; Omar observe aujourd'hui le reflux. Plusieurs partenaires du lancement de Gemma 4 avaient prévu des fine-tunes du modèle de vision 27B et ont annulé en cours de route, car le modèle de base suffisait déjà. Les ajustements comportementaux généraux qui nécessitaient autrefois du fine-tuning passent maintenant par le prompting. Ce qui reste : le fine-tuning de domaine pour la santé, la finance et les données de niche — plus le défi organisationnel de gérer la compatibilité LoRA quand le modèle de base est mis à jour. > *"J'observais beaucoup ces phénomènes — et aujourd'hui je constate moins d'enthousiasme pour le fine-tuning appliqué aux modèles conversationnels généralistes."* ## [16:29] Arbitrages entre architectures denses et creuses Gemma 4 propose deux grands modèles à des comptes de paramètres similaires : un dense à 31B (intelligence brute maximale, tient sur un GPU grand public une fois quantisé) et un MoE à 27B avec 4B paramètres actifs (inférence la plus rapide dans la même enveloppe matérielle). Ces choix de taille ont été délibérément pensés pour les développeurs. L'avertissement d'Omar aux praticiens du fine-tuning : les recettes d'entraînement et les hyperparamètres MoE ne se transfèrent pas proprement depuis les modèles denses — le changement de distribution frappe le routage de façon mal comprise, peut-être parce que les modifications de la distribution d'entrée changent quels experts s'activent. > *"Les MoE sont difficiles à fine-tuner. Ils fonctionnent très bien pour l'inférence, mais en fine-tuning, les gens se heurtent à des difficultés."* ## [18:29] Intelligence par paramètre et recherches futures Sur Gemma 2, 3 et 4, Google a maintenu le nombre total de paramètres à environ 30B tout en rehaussant sensiblement le plafond de capacité — démonstration directe de l'amélioration de l'intelligence par paramètre. Le problème de comparaison devient plus difficile : dès qu'on introduit la sparsité MoE et le déchargement de paramètres, les comptes de paramètres cessent d'être une monnaie commune. L'horizon honnête d'Omar : les limites en connaissance sont probablement structurelles — un modèle à 30B dans trois ans manquera toujours les faits très pointus, car la théorie de l'information limite la compression possible dans un poids fixe. > *"Quelle est l'intelligence par paramètre ? Comment maximiser cette intelligence par paramètre ?"* ## [20:09] Gemma Scope et interprétabilité mécaniste Google a publié Gemma Scope en décembre — un outil d'analyse des activations par couche sur les modèles Gemma 3, appuyé par un jeu de données d'activations de plusieurs téraoctets (peut-être à l'échelle du pétaoctet) couvrant toutes les couches. Omar présente l'interprétabilité mécaniste comme une rampe d'entrée en ML research à faible coût de calcul : pas besoin d'un cluster d'entraînement pour faire de l'analyse d'activations, et les expériences donnent une intuition concrète sur le fonctionnement interne des transformers. > *"C'est un domaine où vous n'avez pas besoin de beaucoup de calcul pour commencer. Cela vous permet de comprendre comment un modèle fonctionne."* ## [21:12] À l'intersection de la recherche et de l'ingénierie Ce qui a motivé la venue des chercheurs à une conférence d'ingénieurs : les ingénieurs font davantage confiance aux modèles quand ils comprennent comment ils ont été construits, même s'ils ne les entraîneront jamais eux-mêmes. Omar et swyx notent tous deux que la frontière entre recherche et ingénierie s'est brouillée — la plupart des travaux de recherche sont des ablations empiriques plus proches de l'ingénierie que de la théorie, et les agents de codage donnent aux ingénieurs un accès direct à l'expérimentation qui nécessitait auparavant un bagage de chercheur. Omar cite la communauté franken-merge et Axolotl comme exemple de Reddit et Discord redécouvrant indépendamment des techniques que les labos de recherche ont ensuite publiées en articles. > *"Il y a beaucoup d'expérimentation empirique — voir ce qui marche ou non, déplacer des éléments — ce qui pour moi relève bien plus de l'ingénierie que de la recherche."* ## [23:59] Regards sur l'« auto-recherche » et l'automatisation agentique swyx pose la vraie question : l'auto-recherche se réduit-elle à des « sweeps de paramètres agentiques », ou peut-elle produire des découvertes à la Move 37 que personne n'aurait cherchées ? Omar reste prudemment sceptique — le bilan d'AutoML était surtout de la recherche par grille déguisée, et le travail d'architecture en profondeur est probablement non automatisable dans les un à deux prochaines années. En revanche, il pense que le fine-tuning lui-même sera bientôt entièrement piloté par des agents : les utilisateurs demanderont à un agent de lancer des expériences plutôt que d'écrire du code d'entraînement, via des outils comme AutoTrain de Hugging Face ou le CLI d'Axolotl. > *"La prochaine génération de praticiens du fine-tuning sera composée de gens qui ne coderont pas du tout. La plupart affineront avec quelques instructions."* ## [26:06] Expansion de l'équipe, hubs mondiaux et intégration de Kaggle L'équipe DevEx recrute maintenant à Singapour et en Inde — co-localisée avec les bureaux de recherche de DeepMind, pour que les DevRel puissent aller voir les chercheurs dans le couloir plutôt que d'être isolés dans des antennes commerciales. La grande nouvelle côté organisation : Kaggle a rejoint DeepMind, et son infrastructure de compétitions et de benchmarks se connecte directement aux lacunes de capacité de Gemma/Gemini — les benchmarks créés par la communauté peuvent remonter comme signal d'entraînement. Omar décrit le modèle comme une boucle de feedback : l'équipe s'engage sur les réseaux sociaux et lors d'événements pour comprendre ce que construisent les développeurs, puis rapporte ce signal côté modélisation. > *"La façon dont nous développons Gemma, Gemini et tous nos outils repose vraiment sur les retours des startups, de la communauté, des développeurs."* ## Entités - **Omar Sanseviero** (Personne) : Responsable Developer Experience chez Google DeepMind ; a auparavant développé le DevRel chez Hugging Face ; dirige l'écosystème développeur de Gemma. - **swyx** (Personne) : Animateur du podcast Latent Space ; intervieweur à AI Engineer London 2026. - **Gemma 4** (Logiciel) : Famille de modèles open de Google intégrant une architecture d'embeddings par couche (déchargement de paramètres effectifs E2B), variantes 2B/4B/27B MoE/31B dense, support de 140 langues, entrées multimodales. - **Gemini Nano** (Logiciel) : Modèle on-device construit sur l'architecture Gemma, embarqué dans les téléphones Pixel et Samsung haut de gamme via l'OS. - **Gemma Scope** (Logiciel) : Outil de Google pour l'interprétabilité mécaniste — analyse les activations par couche des modèles Gemma 3 ; publié en décembre 2025 avec des données d'activation à l'échelle du pétaoctet. - **Gemini Diffusion** (Logiciel) : Transformer de diffusion expérimental de Google pour la génération de texte (pas d'images), annoncé à Google I/O ; principal avantage : la vitesse d'inférence. - **Kaggle** (Organisation) : Plateforme de compétitions et de benchmarks ayant rejoint Google DeepMind ; intègre les évaluations communautaires dans la boucle de feedback des capacités de Gemini. - **Google DeepMind** (Organisation) : Laboratoire d'IA consolidé de Google ; périmètre couvrant Gemma, Gemini, la robotique, le ML on-device et l'interprétabilité mécaniste. - **AI Engineer London** (Organisation) : Conférence d'ingénierie en IA appliquée (édition 2026) ; lieu de cet entretien, ville natale de DeepMind. - **MoE (Mixture of Experts)** (Concept) : Architecture creuse où seul un sous-ensemble de paramètres s'active par token ; inférence plus rapide qu'un modèle dense à paramètres équivalents, mais plus difficile à fine-tuner en raison du routage sensible à la distribution. - **Embedding par couche** (Concept) : Innovation architecturale de Gemma 4 — une table d'embeddings insérée à chaque couche du transformer, permettant à 3 milliards de paramètres de résider hors GPU sans coût de multiplication matricielle. - **Intelligence par paramètre** (Concept) : Le ratio capacité/poids que Gemma 2→3→4 a amélioré en maintenant le nombre total de paramètres constant à environ 30B.

Ivan Burazin, CEO of Daytona, discusses the massive shift from building developer environments for humans to providing composable computers for AI agents. With 74% month-over-month growth and 850,000 daily runs, Daytona provides the bare-metal infrastructure required for stateful, high-performance agentic workflows. This conversation explores the technical challenges of spiky compute, the $10 trillion computer-use market, and why the future AI cloud will look more like Stripe than AWS. ## [00:00] Hook Ivan Burazin describes the intense, direct demand for Daytona's infrastructure, with potential users calling him personally to request access. This level of interest signaled a massive, untapped market for providing execution environments to every future AI agent. The team realized they had identified a critical missing piece in the AI development stack. > *I've never experienced this that people literally call you if you do not give them access. Like they want access right now.* > *[0, 0]* > * ] }, { * > *title": "Introduction* > *{'start': 72.0, 'summary': "Host swyx introduces Ivan Burazin, noting their shared history in the developer experience and 'end of localhost' movements. Ivan recalls reaching out to swyx years ago for advice on developer experience while working at a previous role. They reflect on how their early interactions and mutual interests in cloud-based development tools eventually led to their current collaboration.", 'quotes': ['I was one of the co-founders of code anywhere... we were thinking a long time of like local host should die.', [1, 36], '\n ]\n },\n {\n ', 'title": "CodeAnywhere', 'Shift', 'and the end of localhost', {'start': 195.0, 'summary': 'Ivan discusses his long history with his co-founder, dating back to early 2000s virtualization and the creation of CodeAnywhere. As the first browser-based IDE, CodeAnywhere predated modern infrastructure like Docker and Kubernetes, which provided the team with deep foundational knowledge. After a successful run with the Shift developer conference, they returned to their infrastructure roots to launch Daytona.', 'quotes': ['We originally started stacking stacking servers doing like virtualization in the early 2000s... and that was a services company which we sold.', [3, 38], '\n ]\n },\n {\n "title": "What Daytona is: composable computers for AI agents",\n "start": 358.0,\n "summary": ', "Ivan defines Daytona as a provider of 'composable computers' for AI agents", "moving beyond the limited industry term 'sandboxes.' He explains that agents require diverse computing environments tailored to specific tasks", 'much like different hardware setups for human professionals. This API-driven infrastructure allows agents to execute code in production-grade environments rather than just temporary test boxes.', {'quotes': ['What Daytona is today is essentially composable computers for AI agents... the market calls them sandboxes which [is] misleading.', [6, 41], '\n ]\n },\n {\n ', 'title": "The pivot from dev environments to AI sandboxes', {'start': 487.0, 'summary': "Ivan explains how observing early agents like Devon and OpenHands led to a realization that AI agents require a dedicated compute runtime. While their initial SaaS offering for human automation saw low traction, it attracted developers who specifically needed sandboxes for their agents. This feedback loop revealed a massive, underserved market for agent-specific infrastructure that standard cloud providers weren't addressing.", 'quotes': ['a lot of people reached out that were building agents and they were like hey my agent needs a compute sandbox runtime', [8, 50], '\n ]\n },\n {\n ', 'title": "The New Year’s Eve MVP and customers begging for API keys', {'start': 617.0, 'summary': "On New Year's Eve, Ivan 'vibe-coded' the first MVP of what would become the new Daytona. Although the CTO initially dismissed the code as 'garbage,' the core idea was strong enough to warrant a two-week professional rebuild. When they demoed this version to previous skeptics, the response was immediate and overwhelming, with users demanding API access before the calls even ended.", 'quotes': ["I've never experienced this that people literally call you if you do not give them access.", [12, 18], '\n ]\n },\n {\n ', 'title": "Bare metal', 'stateful sandboxes', 'and Daytona’s scheduler', {'start': 776.0, 'summary': "The team approached the technical architecture from first principles, deciding to run on bare metal rather than traditional VMs. They aimed to combine the speed of AWS Lambda with the stateful, long-running nature of an EC2 instance. This allows agents to 'pause and come back' to their work, much like a human closing a laptop lid, without losing state or performance.", 'quotes': ["agents will be like humans in the sense of you don't want your laptop to be shut down until you're done with work", [13, 57], '\n ]\n },\n {\n ', 'title": "60ms startup', 50, 0, 'sandboxes', 'and 850K daily runs', {'start': 1048.0, 'summary': "Daytona's infrastructure is optimized for both individual speed and massive concurrency, with a single instance spinning up in just 60 milliseconds. This scale supports high-volume customers who perform nearly 850,000 runs daily, with some requesting capacity for half a million concurrent CPUs. The system utilizes a custom scheduler and local NVMe drives to eliminate network latency and maximize IOPS.", 'quotes': ['Our time to spin up one is 60 milliseconds with network latency... if you want to spin up 50,000 at once, we are now at about 75 seconds.', [17, 40], ',\n ', 'The biggest customer of ours does like about 850', 0, "every single day is sort of where they're where they're just shy of a million.", [18, 17], '\n ]\n },\n {\n ', 'title": "Spiky RL/eval workloads and the new agent infra problem', {'start': 1313.0, 'summary': "The 'spiky' nature of AI workloads presents a major challenge for compute providers, leading to a mean utilization rate of only 15% despite peaks hitting 90%. Workloads are categorized into 'background agents' that follow human cycles and 'evaluations/RL' which fire off massive bursts of activity at unpredictable hours. To manage this, Daytona must use capacity commits to handle sudden bursts of 100,000 or more CPUs.", 'quotes': ["Daytona's mean utilization is 15%... because it's very spiky. But it's very spiky but we get up to 90%.", [23, 1], '\n ]\n },\n {\n ', 'title": "RL workloads', 'Kubernetes pain', 'and dynamic resizing', {'start': 1692.0, 'summary': "Daytona competes primarily against managed Kubernetes services like EKS and GKS, positioning itself as a more ergonomic 'Twilio or Stripe' for compute. Unlike Kubernetes, Daytona offers a seamless API for spinning up sandboxes with significantly faster startup times. A key advantage is the ability to dynamically resize sandboxes on the fly to prevent out-of-memory (OOM) errors, a feature difficult to implement on other platforms.", 'quotes': ["Daytona although it's a compute provider it's more akin to a Twilio and Stripe from a consumption perspective than it is an AWS", [29, 46], '\n ]\n },\n {\n ', 'title": "Why every AI agent needs a computer', {'start': 2011.0, 'summary': "Ivan outlines the massive scale of knowledge work, estimating a $50 trillion global salary pool, much of which is locked in legacy Windows applications. He argues that true automation requires 'human emulators' that can interact with these legacy systems via GUIs when APIs are incomplete. By automating 40% of this work, the market opportunity for agentic computer use reaches approximately $10 trillion annually.", 'quotes': ['If you take 40% of that, you get to essentially like 10 trillion dollars a year.', [35, 20], '\n ]\n },\n {\n ', 'title": "macOS sandboxes and Apple’s licensing problem', {'start': 2328.0, 'summary': "The discussion shifts to the difficulties of hosting Mac OS sandboxes compared to Windows and Linux. Apple's restrictive licensing only allows two parallel VMs per machine and requires a 24-hour lock-in for users, making per-second billing economically unfeasible. Furthermore, security restrictions prevent moving memory snapshots between physical machines, severely limiting the scalability of agentic workloads on Mac hardware.", 'quotes': ['Apple is shooting itself in the foot... if it would just enable a concurrency model similar to what you can get on a Windows.', [40, 52], '\n ]\n },\n {\n ', 'title": "Why CLI may matter more than MCP', {'start': 2668.0, 'summary': "The discussion compares the Model Context Protocol (MCP) to the Command Line Interface (CLI) for agentic action. While MCP acts as an interface for APIs, the CLI allows agents to execute scripts and perform deep data analysis within a sandbox. This layer of indirection enables more complex agentic workflows beyond simple data retrieval, allowing agents to actually 'do things' rather than just integrate.", 'quotes': ['the MCP is an interface against an API whereas the CLI is like you can actually go do things... the difference between integrations and actually running scripts.', [45, 34], '\n ]\n },\n {\n ', 'title": "Open source', 'GitHub stars', 'and agent integration', {'start': 2891.0, 'summary': "Ivan details Daytona's transition to an AGPLv3 license for its sandbox product to balance openness with commercial protection. This 'copyleft' approach allows enterprise use but prevents competitors from building proprietary forks without contributing back. Keeping the core engine transparent builds trust with users and allows large enterprises to bypass lengthy security audits by providing agents with full context.", 'quotes': ["in the new sandbox product we did add a AGPL3... you essentially can't make a competitor without open sourcing your stuff.", [49, 49], '\n ]\n },\n {\n ', 'title": "Git', 'CI/CD', 'and agent collaboration bottlenecks', {'start': 3191.0, 'summary': 'Current versioning systems like GitHub are often too slow for the high-velocity output of AI agents, leading to bottlenecks in CI/CD pipelines. Some developers are creating makeshift solutions like dumping codebases into JSON files on S3 to bypass Git overhead. There is a growing need for an agent collaboration layer that precedes the traditional Git-based pipeline to handle companies generating over 1,000 PRs per day.', 'quotes': ["GitHub as-is was an overhead... it wasn't fast enough what they needed.", [54, 3], '\n ]\n },\n {\n ', 'title": "Founder life and building a 25-person infra company', {'start': 3495.0, 'summary': "Daytona's success stems from a core team of 13 people who have worked together for over seven years, fostering a high-trust culture. Ivan acknowledges the difficulty of the founder journey, including being away from family, but posits that growth requires 'pain.' He views his work as building the spiritual successor to serverless and Kubernetes for the agent era, requiring radical responsiveness as a differentiator.", 'quotes': ['Of the 25 people in Daytona, I think about 13 of them we have worked with seven years plus.', [58, 57], '\n ]\n },\n {\n ', 'title": "AI SaaS', 'token resale', 'and API-first business models', {'start': 3764.0, 'summary': 'Ivan presents a critical take on the SaaS ecosystem, arguing that the market is incorrectly applying a premium to vendors who simply resell AI tokens. He points out that these models have significantly worse margins than traditional SaaS. Instead, he advocates for companies to expose their data via APIs and charge for consumption, allowing for actual revenue acceleration through increased agentic usage.', 'quotes': ["The market is adding premium to SAS vendors that are reselling tokens. And I think that's incorrect.", [62, 54], '\n ]\n },\n {\n "title": ', 'GPU sandboxes', 'data centers', 'and compute growth', {'start': 3970.0, 'summary': 'Daytona plans to introduce GPU sandboxes to support workloads like 3D rendering and reinforcement learning on CAD, rather than focusing on inference. While the company currently runs on bare metal via colocation providers, Ivan notes they are architected to potentially own data centers in the future. He currently avoids the high capital risk of building data centers for single-digit margin gains.', 'quotes': ['We will [offer GPUs], but not for inference. Like essentially what we think about is like the GPU sandbox.', [66, 21], '\n ]\n },\n {\n ', 'title": "Why the AI cloud may look more like Stripe than AWS', {'start': 4188.0, 'summary': "The conversation concludes by imagining the 'AWS for AI Agents,' which Ivan suggests might look more like Stripe than a traditional cloud provider. This future 'AI Cloud' will integrate sandboxes, web search, and databases as fundamental primitives. While companies like Cloudflare and OpenAI are competing for this space, Ivan hints that many more infrastructure primitives for agents are yet to be developed.", 'quotes': ["There will be a cloud built out specifically for agents and so that cloud will have sandboxes and it will have web search and it'll have databases.", [70, 47], '\n ]\n },\n {\n ', 'title": "Closing thoughts', {'start': 4286.0, 'summary': 'The discussion ends with the observation that the AI infrastructure market is growing at an unprecedented baseline of 40-75% month-over-month. Ivan and swyx reflect on the race to secure hardware and the shift toward specialized agent clouds that will define the next decade of computing.', 'quotes': ["The entire infrastructure market is growing 40% plus or minus month over month... if you're not growing 40%ish... you don't have to come to work.", [68, 23], '\n ]\n }\n ],\n ', 'entities": [\n {\n "name": "Ivan Burazin', {'type': 'person', 'description': 'CEO of Daytona and co-founder of CodeAnywhere.'}, {'name': 'swyx', 'type': 'person', 'description': 'Host of Latent Space and early investor in Daytona.'}, {'name': 'Daytona', 'type': 'organization', 'description': 'A company providing composable computers and sandboxes for AI agents.'}, {'name': 'CodeAnywhere', 'type': 'organization', 'description': 'The first browser-based IDE, co-founded by Ivan Burazin.'}, {'name': 'Devon', 'type': 'product', 'description': 'An early AI software engineer agent.'}, {'name': 'OpenHands', 'type': 'product', 'description': 'An open-source AI agent project formerly known as OpenDevin.'}, {'name': 'Kubernetes', 'type': 'technology', 'description': "Orchestration technology mentioned as a competitor to Daytona's ergonomic API."}, {'name': 'Apple', 'type': 'organization', 'description': 'Mentioned regarding restrictive Mac OS virtualization licensing.'}, {'name': 'Salesforce', 'type': 'organization', 'description': 'Cloud-based software company mentioned for its API-first strategy.'}, {'name': 'GitHub', 'type': 'organization', 'description': 'Developer platform noted for being a bottleneck in agentic CI/CD workflows.'}, {'name': 'Nvidia', 'type': 'organization', 'description': 'The primary provider of GPUs whose supply constraints dictate market growth.'}, {'name': 'Stripe', 'type': 'organization', 'description': 'Used as a comparison for the consumption-based model of the future AI cloud.'}], 'tags': ['ai-agents', 'infrastructure', 'sandboxing', 'bare-metal', 'cloud-computing', 'developer-tools', 'computer-use', 'saas-growth'], 'seo_title': "AI Agents Need Computers: Ivan Burazin on Daytona's Pivot", 'seo_description': 'Ivan Burazin explains why AI agents need composable computers and how Daytona pivoted from dev environments to 850K daily agent runs.', 'confidence': {'score': 0.98, 'rationale': 'The summary synthesizes multiple detailed chunks covering technical metrics, business strategy, and market philosophy with high fidelity to the source.'}}]}]}]}]}]}]}]}]}]}]}]}]}]}]}]}]}]}]}* ## [01:12] Introduction ## [03:15] CodeAnywhere, Shift, and the end of localhost ## [05:58] What Daytona is: composable computers for AI agents ## [08:07] The pivot from dev environments to AI sandboxes ## [10:17] The New Year’s Eve MVP and customers begging for API keys ## [12:56] Bare metal, stateful sandboxes, and Daytona’s scheduler ## [17:28] 60ms startup, 50,000 sandboxes, and 850K daily runs ## [21:53] Spiky RL/eval workloads and the new agent infra problem ## [28:12] RL workloads, Kubernetes pain, and dynamic resizing ## [33:31] Why every AI agent needs a computer ## [38:48] macOS sandboxes and Apple’s licensing problem ## [44:28] Why CLI may matter more than MCP ## [48:11] Open source, GitHub stars, and agent integration ## [53:11] Git, CI/CD, and agent collaboration bottlenecks ## [58:15] Founder life and building a 25-person infra company ## [1:02:44] AI SaaS, token resale, and API-first business models ## [1:06:10] GPU sandboxes, data centers, and compute growth ## [1:09:48] Why the AI cloud may look more like Stripe than AWS ## [1:11:26] Closing thoughts

Jake Cooper, CEO of Railway, details the platform's evolution from a high-burn startup to a sustainable, bare-metal cloud infrastructure powering 3 million users. He argues that the rise of AI agents necessitates a fundamental rebuild of the cloud, moving away from human-centric tools like Kubernetes and pull requests toward high-density CLI handles and production forking. This conversation provides a roadmap for building modular, high-scale systems capable of supporting the next generation of automated software development. ## [00:00] Intro Jake Cooper argues that developers should stop writing code by hand and instead focus on reviewing agent-generated code to maintain architectural integrity. He emphasizes that while AI tools have improved significantly, underlying architectural patterns matter more than ever in an automated workflow. The hosts introduce Jake as the 'Conductor' of Railway, setting the stage for a discussion on the future of cloud platforms and developer experience. > *you should be reviewing the code that you are writing instead of trying to go and write it by hand.* > *[0, 10]* ## [01:19] What Is Railway? Railway is described as a platform that allows users to deploy applications and databases instantly via a canvas or AI prompts like Claude. Jake explains that the goal is to manage software versioning and environment cloning to reduce the complexity of traditional tools like Docker and Kubernetes. By tracking all changes, Railway enables developers to fork production environments into parallel universes for safe validation without reproducing staging environments manually. > *railway is the easiest way to ship anything.* > *[2, 29]* > *we want to make it really easy for not just to like deploy things, but for you to almost like evolve applications over time.* > *[2, 49]* ## [03:26] Jake’s Path to Railway Jake details his professional journey from front-end work at Wolfram to building distributed systems for Jump bikes at Uber using Cadence. He describes his engineering philosophy as a willingness to 'swim to the bottom of the pool,' which includes writing kernel patches to ensure the best possible user experience. Additionally, he critiques GitHub's architecture, specifically the 'broken pointers' created by cloning, which complicates upstream contributions. > *we will swim to the bottom of the swimming pool to go and get the experience* > *[4, 35]* > *GitHub's original sin is that it's like almost a series of broken pointers.* > *[6, 2]* ## [07:32] Railway’s Six-Year Growth Story Jake presents a growth chart illustrating the rapid increase in daily signups for the Railway platform, which has transitioned from a 'slow grind' to adding 100,000 users weekly. Early growth was driven by high-touch interaction on Discord and a determination to acquire the first 100 core users manually. This visual data serves as a transition into the company's history of scaling and its move toward becoming a primary cloud provider. > *so I just wanted to like pull up this glorious chart you say which is basically your usage or number of daily signups* > *[7, 34]* > *Trying to get those initial like first 100 users to like actually kind of come back to it.* > *[8, 21]* ## [10:11] Rebuilding the Business After the Free Tier At one point, Railway was losing $500,000 a month while only generating $50,000 in revenue, despite having $20 million in the bank. Cooper realized this was an unsustainable business model and chose to prioritize long-term viability over vanity metrics, temporarily closing the free tier to rebuild. The company now maintains a lean team of 35 people, preferring to build automated systems rather than throwing headcount at problems. > *We basically had to kind of close off the the free kind of users for a little while, rebuild the business.* > *[11, 47]* > *We're 35 people right now... we don't want to just like add headcount for the sake of headcount.* > *[10, 52]* ## [12:36] Agents as the Next Software Platform Over the last six months, Railway has prioritized 'agentic' development as the primary mechanism for building and deploying software. Cooper believes the industry is moving from assembly and high-level languages to 'words' as the primary interface. He envisions a future where thousands of agents run in parallel, requiring new tools for coordination and version control to manage the super-exponential growth of workloads. > *We've moved from assembly to C to C++ to JavaScript to now like words.* > *[13, 23]* ## [14:48] Railway’s Infrastructure Philosophy Jake Cooper explains that Railway prioritizes control over low-level primitives like network, compute, and storage to optimize for AI agent workloads. By avoiding Kubernetes in favor of custom orchestration, the team can place workloads with high precision to ensure memory efficiency. This level of control is necessary to prevent cost structures from ballooning as agent usage increases and requires thousands of parallel instances. > *you have to be very very efficient with these agents... or you're going to massively massively blow up your cost structure* > *[15, 10]* > *How do you get agents to coordinate? How do you go and get them to be able to like safely version changes?* > *[14, 28]* ## [17:01] Bare Metal, Cloud Economics, and the Compute Crunch Cooper describes the transition to bare metal as highly lucrative, reporting a payback period of just three months compared to cloud rental costs. This strategy allows the company to achieve 70% margins while leveraging hardware that remains viable for several years. He also notes the surprising appreciation of hardware assets, such as RAM, due to the global compute shortage and supply chain constraints. > *our payback period when we go to to metal... if we rent it in the cloud, our payback period is about 3 months.* > *[17, 2]* > *hardware and all of this stuff is... appreciated in value because RAM has gone up* > *[17, 50]* ## [18:41] Cloud Bursting and Five-Cloud Networking To maintain growth without being compute-constrained, Railway utilizes a hybrid cloud strategy for bursting capacity across AWS, GCP, and Oracle. This required building a custom network overlay capable of straddling five different cloud environments simultaneously. While this complexity led to past reliability challenges, it now allows Railway to scale rapidly regardless of individual provider quotas or hardware availability. > *I spent a weekend rebuilding our entire like network like overlay essentially so that we could straddle uh five different clouds* > *[19, 41]* > *we still maintain like cloud presence for like bursting essentially* > *[18, 52]* ## [21:39] Data Center Debt and Infra Financing Cooper highlights the strategic use of data center debt, secured against hardware, as a more efficient alternative to venture capital for infrastructure expansion. By treating compute capacity as a linear driver of revenue, Railway can scale as quickly as they can deploy new hardware. He encourages infrastructure startups to explore diverse financing tools rather than relying solely on expensive venture equity for physical assets. > *we can scale revenue as basically as quickly as we can scale compute* > *[21, 20]* > *our margins on metal are like quite high for the like 70%.* > *[20, 46]* ## [24:50] Data Centers in Space Jake Cooper and the hosts explore the technical challenges of placing data centers in space, specifically the issue of heat dissipation in a vacuum. Cooper expresses skepticism toward current proposals that ignore fundamental thermodynamic laws, comparing the 'figure it out later' mentality to science fiction. He highlights the difficulty VCs face in distinguishing between visionary ideas and technical 'grifts' in the space-tech sector. > *I haven't seen anybody like prove how you're going to go and dissipate that much heat in a vacuum* > *[25, 16]* > *how do you know what's like basically not possible and like is a grift versus like uh is possible but like sounds completely insane* > *[26, 16]* ## [26:43] What Agents Need From Infrastructure Cooper outlines the infrastructure needs of AI agents, noting they require versioning, observability, and storage similar to humans but at a 1000x scale. He predicts that current industry standards like Kubernetes and Envoy will become bottlenecks as agentic workloads compress development cycles. To support this growth, infrastructure must be modular enough to allow for the rapid replacement of failing components without human intervention. > *the workload profile doesn't change so much as it gets like massively massively compressed because you need to do thousands of these things* > *[28, 28]* > *you just need at a thousandx scale* > *[29, 13]* ## [29:43] CLIs, Canvas, and Agent-Native UX Cooper explains that while humans prefer simplicity, agents benefit from high-density CLI interfaces with numerous flags that serve as 'handles.' The Railway Canvas is also evolving into an output mechanism and 'context anchor' rather than just an input tool. This hierarchical view of infrastructure prevents critical knowledge from being siloed as teams scale complex 'hyperstructures' using automated agents. > *If you hand it to an agent and you say, 'Hey, that's 40 arguments and 600 flags.' Like, oh yeah, this is excellent.* > *[30, 35]* > *It has to be almost like an anchor for your context. It has to be like a port in the storm.* > *[34, 27]* ## [36:34] Central Station, Incidents, and Responsible Disclosure Railway utilizes an internal tool called Central Station to aggregate feedback and user context, moving away from static communication channels like Slack. The team emphasizes transparency by exposing real-time metrics and detailed incident reports, operating under a core value of 'honor.' This approach involves over-disclosing issues to users rather than providing vague or misleading information during outages. > *We'd rather overdisclose and know that you know that something is wrong versus almost like having your provider gaslight you.* > *[40, 22]* > *If you can dynamically aggregate that information and dynamically route it to the right person... this is no longer a manual process.* > *[37, 10]* ## [41:49] Safe Rollouts, SRE Agents, and Production Forks To mitigate the impact of bugs, Railway employs incremental rollouts and makes it easy to test behaviors in safe, shadowed environments. Cooper argues that production should not be treated as 'sacred' to the point of stagnation; instead, infrastructure should allow for trivial production forks. This is essential for AI agents, which face a 'stacking entropy' problem without safe iteration primitives to prevent system drift. > *We've built so much ceremony around like production is sacred... we need to get to a point where it's just trivially easy to test different behaviors.* > *[41, 33]* > *I think if you don't have the primitives to make iterating in production safe, it becomes very very difficult.* > *[44, 3]* ## [46:19] AI SRE, Specs, Code, and Tests Jake Cooper reflects on his transition from an AI skeptic to a believer, noting that the safety of AI SREs depends on infrastructure primitives. He advocates for the 'Holy Trinity' of software engineering: a clear specification, the code, and the tests. By aligning these three, developers and agents can reconcile discrepancies and maintain system integrity during rapid, automated iteration. > *If you just unleash an AI SRE on your production infrastructure... it's going to nuke your production database.* > *[46, 37]* > *You need three points essentially which is you need a clear spec... you need the code and then you need the tests.* > *[48, 22]* ## [49:43] Self-Replicating Infrastructure and the New Serverless The speakers explore the concept of agents using the Railway CLI to modify their own infrastructure, creating a self-replicating loop. This shift necessitates a move away from expensive, static virtual machines toward cheap, instantaneous 'atomic units of deploy' like isolates or sandboxes. The goal is to make throwaway copies of production as trivial and cost-effective as possible for agentic experimentation. > *The agent can like modify its own infra which I think is... yeah it's nuts.* > *[50, 4]* > *How do you go and make those throwaway copies like as trivial as possible to spin up run super cheap etc.* > *[50, 53]* ## [54:37] Heroku, Temporal, and Workflow Engines Cooper attributes the decline of Heroku to Salesforce's lack of focus on compute as a core business, leading to product stagnation. Railway positions itself as a 'fluid compute' provider, leveraging Cooper's decade of experience with Temporal (and its precursor Cadence) for durable workflows. Railway is a power user of Temporal, using it to manage complex, long-running infrastructure tasks at scale. > *The business of Salesforce is to build a really really good CRM... and then you acquire this business as a compute business that's kind of an offshoot* > *[55, 33]* > *I have used Temporal for almost like 10 years now, right? Because like Cadence, all of us other things.* > *[60, 5]* ## [1:05:26] Railpack, Nixpacks, and Lazy-Loaded Filesystems Railway is developing Railpack, an engine for determining source code dependencies, which evolved from their earlier Nix-based tool, Nixpacks. While Nix offers theoretical benefits for versioning, Railway found it caused significant image bloat and scaling issues for real-world workloads. They are now exploring content-addressable file systems to enable lazy loading of data into memory for faster deployments. > *If you want version X and version Y, you end up bloating a lot of your kind of like package like space.* > *[66, 2]* ## [1:07:20] Coding Agents, Token Spend, and Roadmap Acceleration With a monthly cloud spend reaching $300,000, Railway heavily incentivizes the use of AI coding agents among its employees. Cooper argues that manual code generation is an inefficient use of time, urging developers to focus on architectural patterns and code review. This allows the team to 'speedrun' their product roadmap by automating complex infrastructure tasks and test generation. > *If you are writing code by hand you are doing this wrong... you should be reviewing the code that you are writing.* > *[67, 37]* > *If you're not using the AI systems to almost like speedrun your road map... then you're kind of missing a large point.* > *[69, 12]* ## [1:12:15] The Pull Request Is Dying The traditional SDLC is undergoing a radical transformation where the pull request and manual code review are losing relevance. Impact is increasingly measured by the 'percentage of tokens that end up in production' rather than lines of code. As AI systems handle more reconciliation and validation, the focus shifts from the PR to the initial prompt and final deployment. > *The pull request is dying... it's going to be the prompt... and beyond that code review is also kind of dying.* > *[72, 23]* > *The really naive way to go in and measure this is almost like your percentage of tokens that end up in production.* > *[71, 40]* ## [1:13:47] Feature Flags and the Agent-Era SDLC Jake Cooper discusses the critical role of feature flagging in managing the 1000x compression of the SDLC driven by AI agents. He argues that incremental rollouts and blast radius management through flagging will become even more essential for safety as deployment speed increases. This culture of flagging allows for rapid experimentation without compromising system stability for enterprise customers. > *Everything's just going to get compressed by like a thousandx so that everybody can go and do that.* > *[77, 21]* ## [1:17:34] Cattle, Pets, and Cloning Machines Jake offers a contrarian view on the 'cattle not pets' philosophy, suggesting that snapshotting allows developers to treat infrastructure like 'pets' again. By snapshotting every frame and lazily loading file systems, the overhead of traditional DevOps tools like Dockerfiles is reduced. Railway even modifies the kernel to support persistent connections during these system snapshots. > *I think you can move towards having pets so long as... you have a cloning machine for your pets.* > *[78, 2]* > *If you can snapshot every single thing at every frame, then like it actually doesn't matter if you know that obliterated.* > *[78, 12]* ## [1:20:48] Solo Founder Lessons Jake reflects on his path as a solo founder, contrasting it with the Silicon Valley consensus of finding a co-founder. He emphasizes the need to be obsessed with every layer of the stack, from kernel-level changes to go-to-market strategies. He argues that having two co-founders can often lead to deadlocks without a clear tiebreak, whereas solo leadership allows for singular vision. > *Two is the worst number of co-founders is because you have no tiebreak... you basically are like, well, I disagree on this thing.* > *[82, 49]* ## [1:25:31] Focus, GPUs, and Building a New Cloud Railway is intentionally avoiding the GPU provider market for now to maintain its core mission, though Cooper admits GPUs are an inevitable part of their long-term roadmap. He stresses that companies are defined as much by what they choose not to do as by what they execute. The ultimate goal is full vertical integration to ensure a seamless experience from logic to execution. > *I think you're you're defined almost more by the things that you don't do than the things that you do* > *[86, 8]* > *I can tell you for a fact that we will not be doing GPUs now, but we 100% will be doing GPUs at some point.* > *[86, 50]* ## [1:29:39] Closing Thoughts Cooper reveals that Railway is moving toward 100% ownership of its data centers to avoid copying the infrastructure of legacy hyperscalers. By inventing their own infrastructure from scratch, Railway aims to support 'vibe coding,' where the friction between a thought and a live application is completely removed. This approach empowers a new generation of 'citizen developers' to build at the speed of thought. > *there should be no friction in between what your thought is and reality that kind of comes out.* > *[89, 4]* > *we've been very very deliberate to like invent our own infrastructure from scratch.* > *[88, 30]* ## Entities - **Jake Cooper** (person): CEO and 'Conductor' of Railway. - **Railway** (organization): A cloud platform designed for easy deployment and environment management. - **Uber** (organization): Jake's former employer where he worked on distributed systems for Jump bikes. - **Temporal** (software): A workflow orchestration platform used by Railway for reliable infrastructure tasks. - **Salesforce** (organization): The CRM company that acquired Heroku, leading to its perceived stagnation. - **Heroku** (organization): A pioneer PaaS platform that Railway is often compared to. - **AWS** (organization): Amazon Web Services, used by Railway for hybrid cloud bursting. - **GCP** (organization): Google Cloud Platform, one of the five clouds Railway straddles. - **Claude** (software): An AI model mentioned as an interface for deploying on Railway. - **GitHub** (organization): A code hosting platform discussed regarding its architectural flaws in versioning. - **Kubernetes** (software): An orchestration system Railway chooses to avoid for higher-order control. - **Central Station** (product): Railway's internal tool for aggregating user context and support feedback.

Ukraine produced 4 million FPV drones last year; China could produce 4 billion. That asymmetry frames two hours of unusually concrete conversation between Yaroslav Azhnyuk — serial tech founder turned AI-drone builder at The Fourth Law — and economist Noah Smith, who has been writing about the economics of drone warfare since before most Western policy circles took it seriously. They cover the full tech stack (cameras, autonomy modules, fiber optic links, interceptors, a semiconductor fab under construction), a five-level autonomy taxonomy, an eight-dimension autonomous-battlefield framework, and China's manufacturing edge that has no near-term Western answer. The through-line: the West is still planning to fight the last war, Ukraine is the defense valley where the next war is already live, and the gap is widening faster than most people realize. ## [00:00] Cold Open: China's 4 Billion Drones and the Cameras-to-Explosives Pipeline Yaroslav opens cold with a single arithmetic comparison that structures the rest of the episode. Ukraine, not an industrial powerhouse, built 4 million FPV drones in a year. China, with an order-of-magnitude larger manufacturing base and a consumer electronics supply chain already producing the same cameras, motors, and chips, could produce 4 billion. Noah immediately asks whether that makes China the supreme conventional military power on earth right now. Yaroslav won't claim certainty, but won't rule it out either. > *"I don't think we have all the information to claim that, but we cannot count it out. And that alone should be, you know, a big warning sign."* The cold open also plants the personal pivot that the rest of the episode unpacks: Yaroslav went from making cameras that fling treats to pets to cameras that fling explosives to occupiers. ## [01:04] Introduction: Brandon, Noah Smith, and Yaroslav Azhnyuk Guest host Brandon normally runs a science podcast; this episode is the exception. Noah Smith — Noahpinion Substack, economist focused on industrial policy and geopolitics — is co-host and co-interviewer. Yaroslav sets the personal context: on February 23rd, 2022, he and his then-fiancée landed in Kyiv at 11 p.m. on what turned out to be one of the last flights into the city. Eight hours later, the bombs fell. The 17-hour drive west that followed — empty streets, gas stations out of fuel, pouring diesel into windshield-washer canisters — reads like a scene from an apocalyptic film because, for the people living it, it was exactly that. > *"We basically packed our belongings and got in the car and spent 17 hours riding west. That was exactly like that. I, you know, missiles are falling, like there was smoke in Kyiv."* ## [05:41] From Tech Entrepreneur to Defense: PetCube, Brave One, and the D3 Fund Yaroslav's path from pet-tech to defense wasn't a straight line. In San Francisco from 2014 to 2020 building PetCube (one of the leading pet-camera companies), he had never taken military coursework and considered wars a thing of the past. Day one of the invasion he knew he would fight back with everything he could — but weapons weren't the first instinct. Early efforts included lobbying U.S. Congress on Lend-Lease (passed May 2022, underdelivered), co-founding Brave 1 (Ukraine's defense-innovation cluster, analogous to DIU), and helping seed the D3 Fund co-started by Eric Schmidt. By 2023, two things became undeniable: the war would last, and drones had permanently redefined warfare — the first software-defined weapon platform in history, where a battlefield capability upgrade can be pushed overnight like a software update. > *"It's like if you were able to push a software update and get all of your Roman legionaries a new helmet. That has never been possible before."* ## [10:42] The Ethics of Building Weapons: Dual-Use Technology and the Wolf at the Door Brandon raises the dual-use problem: the technology won't stay in Ukrainian hands. Yaroslav's answer is pragmatic rather than philosophical. Every technology from fire to large language models is dual-use; the question for a maker is whether the marginal risk of their contribution outweighs the immediate need. Ukraine is in a forest with a wolf. You deal with the wolf first, then consult Greenpeace. He's clear-eyed that no technology stays contained — the parallel concern about LLMs freely available in North Korea and Russia applies equally to drone autonomy — but frames his own company's responsibility narrowly: they supply to the Ukrainian government and armed forces, not to arbitrary buyers. > *"When you're in a situation where you're in a forest in front of a wolf, you know, you first going to deal with a wolf that wants to eat you and then you're going to go consult Greenpeace."* ## [14:01] The Tech Stack: Cameras, Autonomy Modules, Interceptors, and a Semiconductor Fab The Fourth Law's structure is three interlocking business units. Cameras (daytime and thermal, sold to 200+ Ukrainian drone manufacturers). Drone autonomy modules (sold to the same ecosystem). And UAV products sold direct to the armed forces: FPV strike drones, bombers, Shahed interceptors, and ISR interceptors — drones that hunt Russian reconnaissance drones before they can relay targeting data. The thermal-camera arm is about to start construction on two semiconductor fabs to manufacture sensor chips in-house, driven by the realization that dependence on foreign sensor supply chains is a strategic vulnerability. > *"We're about to start construction of two semiconductor plants to make sensors for thermal cameras. That's super exciting for me as a computer science guy — doing semiconductor, super cool."* ## [18:47] Fiber Optic vs. AI: The Radio Horizon Problem and $32/km Cable The chapter is really about why radio-only FPV drones fail at long range — not just from jamming, but from the curvature of the Earth. Below roughly 60-100 meters altitude at 30-40 km range, a drone enters a radio shadow behind hills, forests, or the horizon itself. The pilot loses video and control precisely when closing on a target that is, by definition, on the ground. Fiber optic cable ($32/km, spooled from the drone) solves the shadow problem but adds weight, limits range, and reduces maneuverability. AI fills the gap differently: terminal guidance lets the drone complete the last few hundred meters autonomously even after the radio link breaks. The two approaches aren't mutually exclusive — you can run AI on top of a fiber optic link to command hundreds of drones with fewer operators. > *"If your drone goes low — and usually Russian infantry and vehicles, they're on the ground and you want to hit them, you need to go low — lower you go, maybe you'll get behind a hill or behind a forest, and if you're far enough you'll just get behind the curvature of the Earth."* ## [25:32] FPV Drones: The New God of War — 70–80% of Frontline Casualties Artillery was historically called "the god of war" because it caused 80% of battlefield casualties. On the current Ukrainian front line, 70-80% of casualties are inflicted by FPV drones — the same fraction, a different weapon. Tanks, designed to dominate land warfare for decades, are now routinely destroyed by $400 consumer-grade quadcopters because armor was never built to defend against attacks from directly above. The trajectory follows the same curve as calculators becoming irrelevant once smartphones arrived: not a linear substitution but an exponential displacement where the new technology's influence grows nonlinearly. > *"They used to say that artillery is the god of war because artillery used to cause like 80% of casualties, and now on that ranking FPV drones rule."* ## [28:28] The Five Levels of Drone Autonomy: From Terminal Guidance to Full Autonomy Yaroslav lays out five autonomy levels describing where the field stands and where it's heading. Level 1 is terminal guidance — the drone flies under human control and locks onto a target only in the final seconds. Level 2 is bombing — dropping munitions from altitude without directly ramming a target. Levels 3-4 introduce increasing target-selection and navigation independence: the drone can identify radio-emitting equipment, track vehicles, or navigate through GPS-denied environments. Level 5 is full autonomy — launch-and-forget, no human in the loop for any mission phase. Current battlefield deployment sits mostly at Levels 1-3. The jump to higher levels isn't primarily a technical problem anymore; it's a deployment, doctrine, and trust problem. Human confirmation remains in the loop at every stage involving lethal targeting decisions — for now. > *"Technology progresses and its influence grows nonlinearly. It's all exponential."* ## [41:37] The Eight Dimensions of the Autonomous Battlefield The five autonomy levels describe a single drone's capability. The eight dimensions describe the full battlefield context those drones operate in. Dimension 1: level of autonomy (the five-level scale). Dimension 2: platform type (quadcopter, fixed-wing, missile, naval drone). Dimension 3: environment (day/night, urban/forest/open terrain). Dimension 4: target type (moving vehicle, static structure, radio emitter). Dimension 5: swarm size and coordination. Dimension 6: command-and-control architecture. Dimension 7: sensing modality (optical, thermal, RF). Dimension 8: infrastructure (simulation, data pipelines, security, deployment tooling). Each dimension interacts with every other. A Level-4 autonomous drone performing well in open daylight terrain may fail completely in a forest at night. Battlefield AI systems have to be evaluated across all eight dimensions simultaneously, not just on the single axis of autonomy level. > *"I say dimension because each of them works with another. It's crucial to understand how autonomy evolves in a modern battlefield environment."* ## [45:32] AI Safety and the Morality of Autonomous Weapons Yaroslav's position flips the standard AI-safety framing: in five to ten years, it will be *immoral* to use weapons *without* AI, because human-only weapons produce more collateral damage and friendly fire. He draws the analogy to manually driven cars — once autonomous vehicles are the norm, letting a human drive on a public road becomes the dangerous choice. Noah pushes to the logical endpoint: a Level-6 "AI general" — one large model that ingests all battlefield data and agentically selects targets, with humans reduced to repairing drones. Yaroslav says technically it could be done now. The constraint is deployment and trust, not capability. He references what was publicly described about AI-assisted target designation in the Iran operation: AI surfaces 127 targets, human reviews the list and presses okay. That's already close to an AI general with a rubber-stamp layer. > *"I think 5 to 10 years from now it will be immoral to use weapons without AI because weapons without AI will be more likely to cause collateral damage or unwanted damage."* ## [51:31] The End of the Rifleman? Noah's 2013 Prediction vs. Battlefield Reality Noah revisits a prediction he made in 2013: the rifleman is obsolete, replaced by standoff weapons. Ukraine both confirms and complicates it. FPV drones have unquestionably displaced the rifle as the primary instrument of attrition — but infantrymen haven't disappeared. They dig trenches, hold terrain, conduct logistics, and survive for months in dugouts under continuous drone threat by adapting: better camouflage, smaller movement signatures, drone-awareness drills. Yaroslav extends the timeline question to humanoid robots. The world is built for bipedal humans; there's genuine utility in a platform that can operate a rifle, open a door, or crew a vehicle. He puts a Terminator-style scenario — humanoid combat robots — at 10 years out, not science fiction. But modern warfare, they agree, is a multi-dimensional problem — dozens of drone types, land ops, reconnaissance, psychological operations, aviation, tanks, logistics — and the press focus on whichever technology is newest understates how much every layer still matters. > *"Modern warfare is really very complex and the fact that drones are the latest coolest thing doesn't mean that now it's that and only that."* ## [01:05:13] China's Manufacturing Advantage and Western Vulnerabilities This is where Noah Smith's economics background drives the conversation. The U.S.-China drone comparison isn't about unit price or autonomy level — it's about manufacturing throughput at scale. China's consumer electronics supply chain already produces the motors, cameras, chips, and battery cells that go into FPV drones. Switching that capacity to military production requires regulatory will, not retooling. Ukraine builds fixed-wing drones with 10 km range from hobby components; China can build fixed-wing drones with 200-300 km range at the same cost curve. The West's vulnerability isn't just quantity. It's thermal cameras (overwhelmingly sourced from China), semiconductor fabs (two generations behind on drone-relevant sensors), and procurement speed (a Western defense contract takes years to award; Ukraine iterates weekly). Yaroslav is optimistic about Western human capital — the engineers exist — but openly frustrated with European institutional inertia and uncertain about whether the U.S. has fully absorbed the lessons from Ukraine and the Middle East. > *"We don't have all the information to claim that, but we cannot count that out. If we want to keep the resemblance of our good past life, we have to do something about it."* ## [01:24:21] Policy Advice for Western Defense: Defense Valley and the Widening Gap Yaroslav's top policy prescriptions are framed around the William Gibson quote he attributes to Arthur C. Clarke: the future is already here, just not evenly distributed. Kyiv is Defense Valley — the place where the future of war arrived first, with hundreds of specialized companies, battle-tested commanders at every rank, and a government that learned to move at startup speed. Priority 1: deep integration with Ukraine's defense ecosystem, not just procurement but embedded learning. Priority 2: procurement reform — the drone-dominance initiative is the right direction and needs to scale 10x. Priority 3: long-range drone readiness for contested maritime environments (Shahed-class drones with 2,000 km range cover the entire Pacific island chain). He worries that the U.S. learned less from Ukraine than it should have and may be repeating the pattern with Iran. > *"Kyiv and Ukraine is sort of the defense valley. It's the point where the future of defense has already arrived, and there's a ton of things to learn from that."* ## [01:32:54] The Drone Race: Who's Ahead, Category by Category Russia was at parity or ahead in drone capability 18 months ago; Ukraine has since pulled ahead on FPV and autonomy. But Russia has a 4x population advantage and significantly more industrial capacity than Ukraine alone — scale disparity is why Western supply matters. The race breaks down by category: FPV strike (Ukraine leads), ISR reconnaissance (contested), glide bombs (Russia leads, dropping from bomber aircraft at scale), deep-strike drones (Russia leads on volume), and interceptors (Ukraine innovating rapidly, Russia catching up). Russia uses helicopters to intercept Ukrainian deep-strike drones — a costly but effective countermeasure revealing how each new offense spawns a tailored defense, at weekly iteration cycles. > *"Everyone says Russia's behind right now in the drone war. But that wasn't true a year ago."* ## [01:41:57] Countermeasures: Shotguns, Jammers, Lasers, and Fishnets Shotguns work — they're the primary kinetic countermeasure against incoming FPV drones — but only for a trained soldier who can hit a 20 cm target moving at 100 km/h under combat stress. Electronic jammers are the most widespread defense: block the radio or GPS link and the drone loses guidance. The catch is that the same spectrum the jammer blankets is often used by your own forces, and jammers are being defeated by frequency-hopping and fiber optic links. Russian tanks now look like porcupines — improvised metal cages and electronic-warfare antennas bolted on top to defeat top-attack drones. Ukraine's answer is shaped charges specifically tuned for the gap between the cage and the hull. Lasers are effective but expensive ($10M+ per system to kill a $400 drone) and slow to slew onto fast-moving targets. Fishnets — literally mesh nets — are being deployed around static positions because they're cheap, snag rotors, and require no power. > *"Then the tanks — if you look at Russian tanks and sometimes Ukrainian tanks or equipment — they all look like porcupines."* ## [01:58:19] The Wedding and Final Takeaway: Be Prepared for War Brandon closes with two questions. First: did Yaroslav actually get married in that chapel on February 23rd? They got legally married, but postponed the reception until the war is over. Second: one takeaway for the audience. Yaroslav's answer is a restatement of the Roman proverb: *si vis pacem, para bellum*. > *"You want peace, be prepared for war. Got to invest in defense and security."* ## Entities - **Yaroslav Azhnyuk** (Person): Founder of The Fourth Law (AI drone autonomy + thermal cameras, Ukraine); previously co-founder of PetCube; co-founder of Brave 1 and D3 Fund; born and raised in Kyiv. - **Noah Smith** (Person): Economist; author of the Noahpinion Substack; co-host for this episode; focus on industrial policy, manufacturing economics, and geopolitics. - **Brandon** (Person): Regular Latent Space host (science podcast background); guest host for this episode. - **The Fourth Law** (Organization): Yaroslav's AI-guided drone company; three business units — thermal cameras, drone autonomy modules, UAV products (FPV strike, bombers, interceptors). Leading drone-AI team in Ukraine. - **PetCube** (Organization): Consumer pet-camera company Yaroslav co-founded in San Francisco (2014–2020); the origin of the "cameras that fling treats / cameras that fling explosives" pivot. - **Brave 1** (Organization): Ukraine's defense-innovation cluster; analogous to DIU (Defense Innovation Unit) in the U.S.; co-founded with Yaroslav's involvement. - **D3 Fund** (Organization): Defense-tech investment fund co-founded with Eric Schmidt (ex-Google CEO) to accelerate Ukraine's drone ecosystem. - **FPV Drone** (Concept): First-Person-View drone — pilot sees through onboard camera in real time; currently responsible for 70-80% of frontline casualties; dominant tactical weapon of the Ukraine conflict. - **Five Levels of Drone Autonomy** (Concept): Yaroslav's taxonomy from terminal guidance (Level 1) to full autonomous operation (Level 5); most current battlefield deployment is Levels 1-3. - **Eight Dimensions of the Autonomous Battlefield** (Concept): Yaroslav's framework for evaluating drone systems across platform type, environment, target class, swarm scale, C2 architecture, sensing modality, and infrastructure. - **Defense Valley** (Concept): Yaroslav's term for Kyiv/Ukraine as the global hub where the future of defense tech is already live — analogous to Silicon Valley for consumer tech. - **Radio Horizon** (Concept): Earth-curvature effect that cuts radio/video links to low-flying FPV drones at 30-40 km range; primary technical driver for fiber optic drone adoption. - **Shahed** (Concept): Iranian-designed loitering munition used by Russia; fixed-wing, up to 2,000 km range; archetype for long-range drone threats to Western bases and Pacific-scenario planning.

Janie Lee et Chai Asawa d'Abridge rejoignent swyx et Jacob Effron de Redpoint pour un croisement Latent Space × Unsupervised Learning sur la façon dont un assistant de transcription médicale est devenu la « couche d'intelligence clinique » de la santé. Ils reviennent sur la philosophie produit inspirée de la climatisation, le cas d'usage de l'autorisation préalable, un système d'évaluation construit autour de cliniciens-chercheurs et de juges LLM, pourquoi HIPAA remodèle le volant de données, et ce qu'implique de traiter 100 millions de conversations médicales de façon fiable. ## [00:00] Introduction L'épisode s'ouvre sur l'argumentaire de Janie Lee : le contexte est tout, les alertes doivent passer du réactif au proactif, et le produit lui-même doit s'effacer comme la climatisation jusqu'à ce qu'un risque clinique justifie une intervention. swyx intervient ensuite brièvement pour inviter les auditeurs à s'abonner plutôt que d'accepter des publicités. > *"L'une des choses que nous aimons dire, c'est que nous voulons que notre produit soit comme la climatisation. Il doit rester en arrière-plan en améliorant simplement les choses."* — Janie Lee ## [01:17] Ce que fait Abridge swyx présente ce croisement annuel Latent Space × Unsupervised Learning, avec Jacob Effron invité en tant qu'investisseur de Redpoint dans Abridge. Janie présente Abridge comme une couche d'intelligence clinique pour les systèmes de santé, née de la documentation : les cliniciens passent 10 à 20 heures par semaine à rédiger des notes, et la conversation patient-clinicien est en amont de presque chaque document produit — la facture, le paiement, le diagnostic. Chai ajoute que tout ce qui se passe avant, pendant et après la visite devient accessible dès lors que l'on dispose d'un contexte complet sur les patients, les payeurs, les recommandations et la littérature. > *"Abridge est une couche d'intelligence clinique pour les systèmes de santé. Nous avons vraiment commencé par la documentation et en construisant pour les cliniciens."* — Janie Lee ## [03:22] De la documentation ambiante à l'intelligence clinique Janie retrace les trois « actes » d'Abridge : gagner du temps (le produit de transcription original qui a rendu leurs soirées aux médecins — le « temps pyjama »), économiser et générer de l'argent pour des systèmes de santé aux marges opérationnelles historiquement basses, et finalement sauver des vies. Le fait que le produit soit ouvert des millions de fois par semaine, avant, pendant et après chaque visite, est ce qui rend l'expansion possible. > *"Ils appellent ça le 'temps pyjama'… des médecins qui rentrent chez eux en pyjama et rattrapent leurs notes chaque soir."* — Janie Lee ## [05:21] Aide à la décision clinique et la primauté du contexte Jacob demande à Chai comment l'aide à la décision clinique d'Abridge se compare à son travail précédent chez Glean. Chai établit le contraste : chez Glean, une mauvaise réponse est ennuyeuse ; en santé, les enjeux sont élevés et la surface utilisateur est bien plus restreinte — moins de personas, mais chaque résultat doit faire mouche. Cela façonne tout, de l'évaluation hors ligne au déploiement progressif, et rejoint la vision façon Jarvis — « un assistant qui vous connaît vraiment » — que chaque hackathon de la dernière décennie a tenté de concrétiser. > *"La vision Jarvis que chaque hackathon auquel j'ai assisté ces dix dernières années proposait toujours un concurrent — mais je pense qu'Abridge est vraiment parti de cette opportunité et continue dans cette direction."* — Chai Asawa ## [08:14] Fatigue des alertes, intelligence proactive et autorisation préalable Jacob soulève le problème classique de la fatigue des alertes : comment décider de rompre le silence de la climatisation et d'interrompre vraiment ? L'exemple concret de Janie est celui de l'autorisation préalable : un refus d'IRM qui arrive aujourd'hui des semaines plus tard peut être transformé en une alerte en temps réel pendant que le patient est encore dans la salle, conditionnée aux politiques des payeurs, aux données EHR, aux diagnostics antérieurs et aux protocoles spécifiques à la clinique. Le frein, c'est la plomberie des données : l'autorisation préalable ne fonctionne que si l'assistant peut assembler chaque signal pertinent au bon moment. > *"Pour rendre possible cet exemple d'autorisation préalable, pensez à toutes les données dont vous avez besoin."* — Janie Lee ## [13:53] Formats de l'IA ambiante et clientèle hospitalière swyx interroge sur les formats. Aujourd'hui la surface principale est le mobile, mais Abridge fonctionne aussi sur desktop, via des plugins navigateur dans l'EHR, sur des dispositifs en chambre pour les environnements d'hospitalisation, les workflows infirmiers, et commence à explorer la RA. Le client est multi-faces : CMIOs, CFOs, CIOs, cliniciens, patients, payeurs et l'industrie pharmaceutique sont tous quelque part dans la boucle, les interactions avec les payeurs se faisant par échange structuré plutôt que par accès direct aux données brutes d'Abridge. > *"Vous parlez beaucoup d'IA ambiante. C'est principalement sur le téléphone ?"* — swyx ## [18:16] Les problèmes d'IA les plus difficiles en santé Interrogé sur le problème d'IA le plus difficile chez Abridge, Chai choisit le support en temps réel de haute qualité, à faible latence et à faible coût dans un environnement clinique à enjeux élevés. Modéliser la longue traîne des politiques des payeurs en représentations intermédiaires sur lesquelles le système peut raisonner en est un exemple précis — la frontière de Pareto ne cesse de bouger, et ils doivent la repousser eux-mêmes plutôt qu'attendre des gains disponibles sur étagère. > *"La frontière de Pareto évolue en permanence, mais nous essayons aussi de le faire maintenant."* — Chai Asawa ## [19:43] Modèles frontier, données propriétaires et stratégie modèle Jacob demande ce qu'ils utilisent tel quel et ce qu'ils construisent. Le cadre de Chai : les modèles frontier absorbent de plus en plus les connaissances médicales générales, si bien que l'avantage d'Abridge réside dans les données propriétaires de conversations médicales et dans les comportements spécifiques à chaque spécialité construits par-dessus. Ils sont explicitement agnostiques quant aux modèles là où c'est possible — ce qui compte, c'est l'expérience produit finale, et ils combinent les solutions selon les workflows. > *"On peut utiliser telle chose pour ceci ou cela — ce qui compte pour nous au bout du compte, c'est la meilleure expérience produit."* — Chai Asawa ## [22:24] L'EHR comme système de fichiers pour les agents La vision de Chai pour l'année à venir : tout agent est au fond un agent de codage, et dans la santé, l'EHR fonctionne comme le système de fichiers — un vaste entrepôt d'informations structurées qui ne tient dans la fenêtre de contexte d'aucun modèle actuel. Janie ajoute que l'objectif reste de garder le clinicien concentré sur le patient : avoir le bon contexte prêt au bon moment, sans ressasser la conversation. > *"Presque tout agent est au fond un agent de codage — on lui donne un système de fichiers, il peut écrire son propre code… On peut penser à l'EHR comme à un système de fichiers."* — Chai Asawa ## [25:20] Personnalisation, mémoire et préférences des cliniciens Jacob demande comment Abridge gère la personnalisation par médecin. La réponse de Janie est en couches : les modifications individuelles deviennent des signaux, les paramètres par défaut propres à chaque spécialité s'ajoutent par-dessus, et les politiques du système de santé enveloppent l'ensemble. Chai parle de la mémoire comme d'un nouveau type de système d'enregistrement — des tâches de fond qui consolident les signaux entre les visites, à l'image du sommeil qui consolide la mémoire chez l'humain, pour que le modèle « apprenne » de chaque modification et de chaque non-modification. > *"L'une des données d'échappement intéressantes pour nous, c'est que la mémoire est en quelque sorte l'un de ces nouveaux systèmes d'enregistrement."* — Chai Asawa ## [31:57] Évaluations, juges LLM et déploiement progressif Janie détaille le système d'évaluation : des cliniciens internes réalisent une première revue « LFD », des juges LLM sont calibrés à partir de ces données annotées, des évaluateurs tiers apportent une lecture indépendante, et des évaluations propres à chaque spécialité captent ce que les évaluations génériques ratent. Chai ajoute une analogie avec les voitures autonomes : ils veulent être au contact de la réalité le plus vite possible, mais uniquement via un déploiement progressif, de sorte que la distribution hors ligne corresponde vraiment à la distribution en production. > *"Je veux être au contact de la réalité le plus vite possible, mais je veux un déploiement progressif car, autant que mes données d'évaluation hors ligne me servent, je veux que leur distribution corresponde à la distribution réelle."* — Chai Asawa ## [38:04] HIPAA, dé-identification et vie privée La confidentialité est traitée comme une contrainte absolue sur le volant de données. Chai explique que toutes les données utilisées comme base d'évaluations en ligne ou d'apprentissage doivent être dé-identifiées de façon irréversible — ils ont conçu des processus autour de cela. Janie ajoute que les contrats clients fixent également qui, au sein d'Abridge, peut accéder aux PHI, si bien que le seuil pour ce qui entre dans les données d'entraînement est élevé contractuellement, pas seulement au niveau des politiques internes. > *"Toutes les données que nous utilisons doivent être dé-identifiées — toute donnée réelle que nous utilisons comme base d'évaluations en ligne ou d'apprentissage, vous devez donc..."* — Chai Asawa ## [40:38] 100M conversations et fonctionnement à grande échelle À 100 millions de conversations et plus, la surface des préoccupations change : le routage des modèles, le post-entraînement, les budgets de fiabilité et le coût par appel deviennent des enjeux de premier plan. Chai parle des insights qu'on peut remonter aux cliniciens, et étend l'horizon temporel : à terme, la même conversation pourrait alimenter des signaux directement vers les patients et les consommateurs, pas seulement les prestataires. > *"Dans notre jeu de données de 100 millions de conversations, on peut imaginer des insights à donner aux cliniciens."* — Chai Asawa ## [45:27] Intégration EHR et la couche d'intelligence clinique swyx interroge sur la relation avec l'EHR. Abridge investit massivement dans une interopérabilité profonde — le partenariat EHR est incontournable pour l'adoption par les cliniciens, mais la valeur qu'Abridge ajoute par-dessus opère à une échelle différente : contexte inter-visites, raisonnement intégrant les payeurs, et ce type d'intelligence clinique que l'EHR lui-même n'est pas structuré pour produire. > *"L'un des partenaires clés est l'EHR, et je me demande à quoi ressemble cette relation."* — swyx ## [47:56] Réglementation santé, latence et IA à enjeux élevés Jacob demande ce qu'Abridge a appris de la réglementation. Janie nuance le discours habituel : l'IA en santé bénéficie en réalité de vents réglementaires favorables, car le niveau d'exigence est si élevé que les problèmes les plus difficiles finissent par être résolus ici en premier. Chai évoque les « astuces ingénieuses » qu'ils déploient aujourd'hui en sachant que la frontière continuera de bouger, et en acceptant que certaines de ces astuces ne survivront pas à cinq ans. > *"Je pense que c'est là que les problèmes d'IA les plus difficiles seront résolus en premier, justement parce que le niveau d'exigence est si élevé."* — Janie Lee ## [51:28] Cliniciens-chercheurs et qualité longue traîne Janie décrit un rôle interne à Abridge appelé le clinicien-chercheur — des médecins qui sont aussi techniques, allant des ingénieurs full-stack aux « prompteurs extrêmement débrouillards ». Les avoir intégrés dans les équipes produit et d'évaluation élève le niveau de ce qui est expédié, car les personnes qui rédigent les critères LFD sont celles qui comprennent vraiment ce que signifie être cliniquement utile. swyx fait le lien avec l'apprentissage actif sur les points faibles connus — ce type de polish qui est un art perdu dans la plupart des équipes IA. > *"Nous avons ce rôle appelé le clinicien-chercheur et j'ai entendu l'un de nos dirigeants les appeler des mutants récemment."* — Janie Lee ## [54:21] Leçons de Glean et infrastructure IA durable Jacob demande à Chai ce qui se transfère de Glean. La réponse porte surtout sur ce qui tient dans le temps — les couches de contexte, les systèmes pilotés par les événements, Kafka, Temporal, les sockets, les CRDTs issus du playbook de collaboration de Google Docs. Les systèmes multi-agents héritent des mêmes problèmes de résolution de conflits que les humains, et les patterns d'infrastructure de la dernière décennie ne sont pas abandonnés — ils sont réutilisés. > *"Il y a beaucoup de technologie pilotée par les événements… que ce soit Kafka, Temporal, les sockets et ainsi de suite — comment tout assembler est aussi quelque chose de durable, je pense."* — Chai Asawa ## [58:20] L'avenir des workflows de santé agentiques Un bref échange sur ce qu'un Abridge plus agentique signifie concrètement : toujours ancré sur le rôle du clinicien dans la relation patient, mais avec davantage de travail en arrière-plan — réagir aux résultats de laboratoire, rédiger des suivis, prendre en charge des capacités au nom du clinicien sans s'emparer de la relation. > *"Encore plus de capacités au nom du clinicien, qui a selon nous un rôle crucial à jouer dans la relation avec le patient."* — Chai Asawa ## [58:51] PRDs, clarté produit et développement de produits IA sérieux La question rapide de Jacob : sur quoi avez-vous changé d'avis en IA au cours de la dernière année ? Janie retourne la tendance populaire : les prototypes ne sont pas la panacée, les PRDs ne sont pas morts. Plus les produits deviennent complexes et propulsés par l'IA, plus la discipline de clarté par l'écrit d'un vrai PRD compte, pas moins. Le reste de la section porte sur la construction de produits IA sérieux en santé : la responsabilité, la discipline de spécification écrite, et la résistance au développement guidé par la démo. > *"La prise de position la plus audacieuse veut que les prototypes soient la seule finalité et que les PRDs soient morts."* — Janie Lee (la position sur laquelle elle a changé d'avis) ## [64:28] Outils de codage IA chez Abridge La question standard de fin d'émission de swyx. Abridge utilise Claude Code et Cursor en interne, et Jacob lance une boutade comme référence — il aimerait voir Claude diriger une entreprise à 1 milliard de dollars avant tout chiffre d'affaires. > *"Claude va faire ça — j'aimerais voir Claude… diriger une entreprise à un milliard de dollars avant tout chiffre d'affaires."* — Jacob Effron ## [65:23] Outro Chai oriente les auditeurs vers le site d'Abridge pour consulter leurs livres blancs — réduction des hallucinations, évaluations, et le reste de la pile de recherche. swyx et Jacob concluent avec des remerciements et des mots de clôture. > *"Sur le site d'Abridge, nous avons beaucoup de nos livres blancs où nous avons réalisé un travail intéressant, comme la réduction des hallucinations."* — Chai Asawa ## Entités - **Janie Lee** (Personne) : Opératrice chez Abridge depuis ses débuts ; responsable des volets produit et commercial de la couche d'intelligence clinique. - **Chai Asawa** (Personne) : Responsable de l'aide à la décision clinique chez Abridge ; anciennement chez Glean. - **swyx** (Personne) : Animateur de Latent Space. - **Jacob Effron** (Personne) : Associé chez Redpoint Ventures ; animateur du podcast Unsupervised Learning. - **Abridge** (Organisation) : Entreprise d'IA en santé construisant la couche d'intelligence clinique — partie de la documentation ambiante, maintenant en expansion vers l'aide à la décision, l'autorisation préalable, les évaluations et l'intégration EHR. - **Glean** (Organisation) : Entreprise de recherche IA en entreprise ; mentionnée comme ancien employeur de Chai et comme point de comparaison horizontal vs vertical avec Abridge. - **Redpoint Ventures** (Organisation) : Fonds de capital-risque ; investisseur d'Abridge et point d'ancrage du croisement Unsupervised Learning. - **EHR (Electronic Health Record)** (Concept) : Le système d'enregistrement sur lequel s'appuient les systèmes de santé ; dans le cadre de Chai, l'EHR fonctionne comme un système de fichiers pour les agents de santé. - **Autorisation préalable** (Concept) : Un cas d'usage central d'Abridge — transformer des refus de payeurs qui prennent des semaines en orientations en temps réel pendant la visite. - **Processus LFD** (Concept) : La revue initiale conduite par des cliniciens internes chez Abridge, utilisée pour calibrer les juges LLM et définir les critères d'évaluation. - **Clinicien-chercheur** (Concept) : Un rôle propre à Abridge — des médecins aussi techniques, intégrés dans les équipes produit et d'évaluation. - **Déploiement progressif** (Concept) : La discipline de déploiement d'Abridge ; mise en production sur une tranche de trafic réel pour maintenir honnête la distribution hors ligne, calquée sur le schéma de mise en production des voitures autonomes. - **Claude Code** (Logiciel) : Outil de codage IA utilisé en interne chez Abridge. - **Cursor** (Logiciel) : Éditeur de codage IA également utilisé en interne chez Abridge.

Matt Pocock joins swyx at AI Engineer Europe to argue that the old software design canon — DDD, deep modules, ubiquitous language — matters more, not less, in the AI coding era. The thesis: code is not just a compile target; a codebase that is easy for humans to change is easy for AI to change. Along the way they cover course-making, why traditional lectures still beat AI-native learning, and TypeScript's quiet takeover of AI engineering. ## [00:04] Opening at AIE Europe and the Cursed Course swyx welcomes Matt to the AI Engineer Europe podcast booth in London. Matt jokes that AIE is "the worst" event he has ever attended (the location is in fact astonishing) before turning to his Claude Code course, which is just wrapping up its two-week cohort. He explains why he runs short cohorts: AI moves so fast that self-paced courses cannot guarantee updates, and the "curse" of releasing into breaking changes — AI SDK v5 dropped on day two of his AI SDK v4 course, and the Claude Code source leaked during this one — is now baked in. The conversation then turns to teaching as a craft. Matt rejects the "pundit" branch of YouTuber identity — he is not trying to predict the future, only to teach durable material — and notes that being a teacher first is what differentiates his content. > *I'm not a guy who's trying to predict the future. I'm just trying to teach.* ## [02:51] Why Engineering Fundamentals Matter More with AI Matt previews his AIE talk. The popular narrative says code no longer matters because English plus an AI compiler can produce applications. Every time he tried to ignore the code, he ended up with "a terrible mess." So he went back to the classics — *Extreme Programming*, *The Pragmatic Programmer*, *A Philosophy of Software Design*, DDD — and discovered they ported directly into prompts. Keeping the architecture in your head, even when you delegate implementation, yields outsized dividends. > *If you have a code base that's easy to change for humans, it's going to be easy for AI to change, too.* ## [04:23] Narrow Waist and Deep Modules swyx introduces the "narrow waist" concept from internet architecture (TCP/IP, HTTP at layers 3–4) as a way to contain AI-generated slop: define rigid interfaces, delegate the inside. He extends it to running AIE as a nine-person business — "model-view-claw" instead of MVC, where coordination across people and AI is the real systems problem. Matt maps this onto John Ousterhout's notion of *deep modules*: a large amount of functionality behind a simple interface, ports and adapters style. This is, in his experience, the best way to use AI for coding — be intentional about the interface as a human, then delegate the implementation. > *Deep modules basically — a large amount of functionality with a simple interface. Kind of ports and adapters, right?* ## [06:37] Domain-Driven Design Meets AI DDD is having a moment, and Matt argues it works *because* the framework has been around long enough to sit in the latent space of these models. You do not have to invent new vocabulary; you can bolt on a system that is composable and that the model already understands. The deeper point: DDD is fundamentally about aligning code with language, which is exactly what you want when speaking to an AI. He makes it concrete with the `mattpocock/skills` repo (≈13k stars) and its "ubiquitous language" skill — a Claude Code skill that scans your codebase, surfaces the arcane jargon, and refines it with you into a markdown file he keeps open while prompting. He references it from `agents.md` but does not paste it wholesale, so the agent finds it when searching for those terms. > *Essentially, you're trying to create a unified domain model so that the AI and you are speaking the same language.* ## [10:05] Teaching as an Overpowered Skill swyx asks how Matt got so good at explaining things. Matt credits six years as a voice coach before becoming a developer — communication felt like an unfair advantage when he started as a junior. He has since narrowed his focus: split time between learning material and finding the right phrases for it. The old texts help because they give him pre-built mental models to explain new ideas through. He walks through his course-making process: an "explore and exploit" phase, a Zettelkasten-style Obsidian vault, a custom planning app, P1/P2/P3 prioritization, and the rule that *each lesson teaches exactly one thing* with dependencies made explicit. Most of what he produces ends up on the cutting room floor. > *The ability to communicate always just felt like a ridiculous overpowered skill that I had in my locker that no one else had.* ## [13:20] How People Actually Learn AI Engineering The conversation turns to whether AI has changed how people learn. Matt distinguishes knowledge (lectures), skills (interactive exercises), and wisdom (small-group discussion — and now, talking to an AI). Counterintuitively, the more he leans into AI-experimental teaching, the more it turns his audience off. Most learners still want traditional lectures; swyx recalls Maven's cohort-based education arc landing in the same place. Matt's compromise is to force the work without forcing the form: in his TypeScript material he throws learners into a problem first and gives them the knowledge afterwards. > *The more I lean into the kind of AI experimental stuff, the more it actually turns people off my materials.* ## [15:04] TypeScript Overtaking Python swyx flags that TypeScript overtook Python in the GitHub survey this year — a shift he did not see coming, particularly in AI engineering where Python's expressiveness has been dominant on the backend. Matt's echo chamber is 100% TypeScript, but his real argument is ecosystem: when you care about UX and shipping chat-style applications, the framework gravity is in TypeScript (Vercel's Next.js, Cloudflare's variants). swyx admits this would meaningfully change which frameworks he promotes. > *If you're concerned about UX, concerned about shipping great stuff, you're mostly doing it in TypeScript.* ## [16:45] Inversion of Control and Composable Skills Matt looks ahead. His TypeScript-evals bet (Everlight) stalled — "no one's excited to do evals." The next frontier is *inversion of control*: as coding agents converge on similar architectures (Firebase-style backends, small tool sets), the interesting axis becomes how much control sits with the developer versus the harness. Claude Code's opacity buys ease of use but loses observability; Pydantic AI ("Pi") swings the other way — total control, total maintenance burden. He closes by pointing past coding agents entirely. Software engineers are a step ahead because AI produces quality output in their domain, but the composable skills he authors — like his three-sentence "grill me" skill that makes the AI interrogate you until you reach a shared understanding — generalize to any domain where you want the AI aligned with you. > *The inversion of control is going to be really important — you put more control in the hands of the developer and less in the harness.* ## Entities - **Matt Pocock** (Person): Creator of Total TypeScript and AI Hero; teaches TypeScript and AI Engineering through two-week cohort courses. - **Shawn Wang / swyx** (Person): Host; founder of AI Engineer and the AIE conference series. - **AI Engineer Europe (AIE)** (Organization): The London conference where this conversation was recorded; Matt's talk hit 1M views in 13 days — fastest in AIE history. - **AI Hero** (Organization): Matt's AI engineering education platform (aihero.dev). - **Claude Code** (Software): Anthropic's coding agent; subject of Matt's just-finished course and a recurring example throughout. - **Domain-Driven Design (DDD)** (Concept): Software methodology centered on aligning code with the language of the business domain; Matt argues it ports cleanly into AI prompting. - **Ubiquitous Language** (Concept): DDD practice of maintaining a shared vocabulary doc; Matt's namesake Claude Code skill scans a repo and refines this with the user. - **Deep Modules / Narrow Waist** (Concept): Architectural pattern (Ousterhout / internet protocols) of large functionality behind a small interface — Matt's preferred shape for AI-assisted codebases. - **mattpocock/skills** (Software): Matt's open-source repository of Claude Code skills; ≈13k stars at recording time. - **Pydantic AI (Pi)** (Software): Python agent framework built from low-level primitives; cited as the high-control counterpoint to Claude Code's opaque harness. - **Obsidian** (Software): Note-taking app reportedly run by a team of four; the example for non-engineering domains where AI leverage compounds.

Alex Lupsasca — 2024 New Horizons Breakthrough Prize winner and OpenAI resident scientist — recounts how GPT-5 resolved a year-long open problem in quantum field theory: proving that single-minus gluon tree amplitudes are non-zero and finding their compact closed form. He then describes how the publicly available GPT Pro, given the gluon paper as a seed, independently generalized the result to graviton amplitudes in under three days of human clock time. Throughout the conversation, Lupsasca reflects on what this trajectory means for how physics is done, how the next generation of physicists will be trained, and where the remaining bottlenecks — verification, creativity, and publishing infrastructure — still lie. ## [00:00] Introduction to AI's impact on physics research Lupsasca opens in medias res, framing the episode's central claim before the formal introduction: AI has crossed a threshold where it can resolve questions that stumped human experts for over a year. He describes this not as a curiosity for theoretical physicists but as a profound, if underappreciated, change in the nature of scientific discovery itself. > *"That's a certain milestone that we've passed, and I think maybe for the average person on the street who doesn't care about theoretical physics, this is not very noticeable, but I think it's a very profound change and we've really passed some kind of a threshold."* ## [00:43] Guest introduction: Alex Luposka The hosts — Brandon (Atomic AI) and RJ Honicky (Miro Omix) — introduce Lupsasca as a Vanderbilt professor and OpenAI fellow who holds both the 2024 New Horizons in Physics Breakthrough Prize (often called the "Oscars for science") and the IUPAP Young Scientist Award. Lupsasca immediately sets the narrative arc: a year ago, AI was useful for email but not for his work; ChatGPT o3 was the first model that genuinely helped with research math; then GPT-5 reproduced one of his hardest published results in 30 minutes. > *"When GPT-5 came out it was able to reproduce one of my best papers that took me a very long time to come up with in like 30 minutes. And that's when I really became AI pilled."* ## [02:49] Alex joining OpenAI and the shift in physics research After GPT-5's release, Lupsasca began evangelizing the shift to colleagues who were skeptical. Finding OpenAI equally excited, and being on sabbatical, he joined as resident scientist — the person physicists around the world now email when something astonishing happens. He describes receiving an inbound that week about Codex simulating the Sachdev-Ye-Kitaev (SYK) model in 10 minutes, a feat that many research groups had struggled to achieve due to the narrow Venn diagram of physicists with strong coding skills. > *"I talked to OpenAI. They were also really excited and I thought I have to get in on this and to understand that this is happening and not be a part of it is a huge mistake so I have to go to OpenAI."* ## [04:08] The release of GPT-5 and the shift in capabilities Lupsasca contrasts the lukewarm Twitter reception of GPT-5 (complaints that it was not better at writing email) with what he observed at the science frontier. He notes GPT-5.4 is another significant jump, and describes how AI capabilities for physics have been accelerating rapidly since o3, the first reasoning model strong enough for research-grade mathematics. He uses this as a bridge to the central technical story of the episode: a pair of new papers on gluon and graviton scattering amplitudes. > *"At the science frontier the capabilities were really taking off."* ## [10:05] Explaining Quantum Field Theory and amplitude calculations Lupsasca gives an accessible primer on quantum field theory (QFT), the framework that reconciles special relativity and quantum mechanics. The key objects in QFT are scattering amplitudes — complex-valued functions that encode the quantum probability for a set of incoming particles (with given energies, momenta, and polarizations) to scatter into a set of outgoing particles. These amplitudes are computed at particle colliders like the LHC, and knowing the n-point amplitude (for any number n of particles) encodes essentially the full content of the theory. > *"If you have a particular force and you're able to compute the n-point amplitudes... you know everything about the theory."* ## [14:20] Overview of gluons and the strong force Gluons are the force-carrying particles of the strong nuclear force — the force that, despite like-charge repulsion between protons, holds the atomic nucleus together. They are the QFT analog of photons for electromagnetism and gravitons for gravity. Like photons, gluons carry a polarization (helicity): positive (right-handed) or negative (left-handed). This helicity structure is central to the paper discussed next. > *"The strong force is mediated by the exchange of the particles of the strong force, which are called gluons, because they're what glues together the nucleus of the atom."* ## [14:38] Discussing the first research paper on single-minus gluon tree amplitudes Lupsasca unpacks the paper's title — "Single-Minus Gluon Tree Amplitudes Are Non-Zero" — piece by piece. Tree amplitudes are the leading-order (no-loop) contributions to scattering. All-plus-helicity amplitudes are exactly zero by a symmetry argument. Single-minus amplitudes — where all but one gluon have positive helicity — were assumed in textbooks to also be zero by the same argument. The paper proves they are not. The result involves collaboration with Alfredo Guevara (IAS), David Skinner (Cambridge), Andrew Strominger (Harvard), and Kevin Wheel. > *"If you look at the lecture notes and textbooks that have been written on this, the same argument that rules out the all-plus amplitudes also appears to rule out the single-minus amplitudes."* ## [20:56] How ChatGPT helped solve a year-long physics puzzle Strominger, Guevara, and Skinner had understood for about a year that the textbook argument has a loophole: when particles are collinear (exactly aligned in momentum), the standard dimensional-analysis reasoning fails, and single-minus amplitudes can be non-zero. But computing what those non-zero amplitudes equal had eluded them. Lupsasca invited Strominger to visit OpenAI and work on it with AI. The week before Strominger's flight, Lupsasca began using ChatGPT Pro. By the time Strominger landed, they had the answer. > *"Using ChatGPT we solved the problem before he even got off the plane."* ## [23:02] Complexity of manual calculations in physics Lupsasca shows the audience a concrete illustration of the difficulty: the six-point single-minus amplitude, worked out by hand by Alfredo Guevara, is a sum of 32 terms each of which is itself a product of four complicated factors. The number of terms grows factorially with the number of particles n — super-exponential growth. This is the messy representation that the group had been staring at for a year, seeking the analog of the elegant Parke-Taylor formula. > *"By the time you get to six terms, it explodes in your face."* ## [26:12] The history and mechanics of Feynman diagrams Feynman diagrams are a visual language introduced by Richard Feynman to organize perturbative QFT calculations: diagrams represent possible intermediate histories of a scattering process, and the full amplitude is a sum over all of them. Diagrams are organized by number of vertices (interaction points); each additional vertex is suppressed by the coupling constant, so tree diagrams (fewest vertices) dominate. Loop diagrams — where intermediate particles are created and annihilated — contribute smaller corrections. The combinatorial explosion of tree diagrams is the root cause of factorial growth. > *"In principle, there are infinitely many pictures to sum over."* ## [27:44] The Parke-Taylor formula and the quest for simplification In the 1980s, Parke and Taylor computed the "maximally helicity violating" (MHV, or double-minus) gluon amplitudes through a heroic Feynman diagram expansion. Despite the factorial number of terms, everything canceled to leave a single compact formula — the Parke-Taylor formula — that fits in half a line. Strominger, Guevara, and Skinner spent a year looking for the analogous compact formula for the single-minus case. Their search stalled at the level of the messy Feynman representation. > *"Andy, Alfredo and David spent the last year chasing the analog of the Parke-Taylor formula, the very simple answer that was obtained in the '80s for the double minus amplitudes."* ## [31:26] Using ChatGPT to find the simplification in the special phase space region When the five-point single-minus amplitude was fed to ChatGPT Pro, the model identified a special subregion of phase space (where one particle's frequency has opposite sign) in which the amplitude simplifies from eight terms to a product of just three. This appears not to have been a known fact; the model wrote Python code and tested thousands of possibilities to deduce it. Moving to the six-point amplitude (Guevara's hand calculation), ChatGPT simplified 32 terms to a product of 4. It then conjectured the general n-point formula — with only linear growth in the number of terms, the best possible behavior. GPT-5.2 Pro guessed the formula but could not prove it. > *"The formula that it proposed, instead of having this factorial growth... here it's actually linear. So if you double the number of particles, you only double the number of terms."* ## [38:07] Proving the formula from scratch to ensure validity To obtain a proof, Lupsasca used an internal OpenAI model with extended reasoning. He gave it the problem cold — without the conjectured formula — and asked it to find the general answer in the special phase-space region. After 12 hours of computation, the model independently rediscovered the same formula and produced a complete three-step proof. The proof constitutes the bulk of the published paper. The team kept the AI attribution to one paragraph, framing the paper as a physics result that stands on its own merits. > *"We gave it the whole problem from scratch... and it came back with the same formula which we had not given it. So it rediscovered the correct formula. But this time it also found the proof."* ## [41:00] Determining the scientific impact and future research Asked to compare the result to the Parke-Taylor formula, Lupsasca is candid that scientific impact is only assessable decades later, but argues the result is genuinely surprising and should open a line of attack toward deeper questions in quantum gravity. The conversation pivots naturally to the second paper. > *"I think the true value of a paper can only be assessed decades into the future based on how much future work it leads to and what developments it opens up."* ## [42:27] Introduction to the second paper on graviton amplitudes Gravitons are the hypothetical quanta of gravity — the spin-2 force carrier analogous to the spin-1 photon (electromagnetism) and gluon (strong force). Unlike gluons, gravitons have never been directly detected, but they are central to quantum gravity theory. The second paper, "Single-Minus Graviton Tree Amplitudes Are Non-Zero," shows the same loophole applies to gravity and that a compact formula extends there too — despite gravitons being mathematically more complex than gluons. > *"We wrote this paper which is called single minus graviton tree amplitudes are non-zero. So it's the same title almost, except with graviton instead of gluon."* ## [45:41] Defining particles, irreducible representations, and symmetry Lupsasca sketches the modern QFT definition of a particle (an irreducible representation of the Poincaré group, classified by Wigner according to mass, spin, and charge) and explains why gravitons are spin-2 while gluons and photons are spin-1, making graviton polarization data twice as rich. Crucially, the second paper was complete within three days of the first going public — most elapsed time was spent verifying correctness, not computing. > *"Most of the time was spent verifying the answer, not writing, which is insane, actually, if you take a step back."* ## [47:46] How GPT Pro generalized the research to gravity For the graviton paper, no internal model was needed — the publicly available ChatGPT GPT-5.2 Pro sufficed. Lupsasca provided the gluon paper as context plus two paragraphs describing the key mathematical changes, then said "Good luck. You're a brilliant theoretical physicist." Over a 110-page exchange, the model worked through the graviton calculation — applying the directed matrix tree theorem, a piece of known combinatorics that neither Lupsasca nor collaborators had thought to invoke — produced correct intermediate results, and wrote a draft paper very close to the final arXiv version from section 3 onward. > *"It's a real solid result in quantum gravity that was done pretty much completely by an AI with human steering it and asking kind of the right questions."* ## [53:57] The epistemological shift: Is this a new way of doing physics? The hosts raise the central epistemological question: if an undergraduate with domain knowledge and good prompting could have done this, what does graduate training mean now? Lupsasca agrees this is the hardest open question facing academia. He notes that arduous calculation trains not just skill but self-confidence, that the gap between coursework and the research frontier is growing, and that many "easy" problems professors once assigned to students are now solvable by AI in minutes. He offers two concrete ways AI has already changed his own workflow: dramatically reducing time spent confused between steps, and enabling parallel AI scouts that explore multiple research directions simultaneously. > *"With AI, actually, you can launch 10 instances of chat and have each one try a different route and send it as a scout that moves very fast into the unknown."* ## [59:27] The use of AI as a 'scout' for research directions Lupsasca elaborates on the scout metaphor: rather than carefully mapping a route from A to C before committing to it, a researcher can now dispatch many AI "scouts" in parallel, get rapid feedback on which directions are promising, and redirect human attention accordingly. Even when a scout makes errors, its signposts reduce orientation cost for the human following. This constitutes a qualitatively new mode of research — one where the bottleneck shifts from calculation to judgment about which directions matter. > *"Even if ChatGPT doesn't always get everything right, just kind of having a scout that signposts some key steps along the way that you can use to anchor your own movement is extremely helpful."* ## [61:44] The role of 'taste' and collaboration with AI The hosts push on the problem of "taste" — the ability to identify which questions are at the productive edge of knowledge. Lupsasca argues that working effectively with ChatGPT requires the same skill a professor develops advising students: knowing what question to give, at what level of detail. "Taste" — knowing where the frontier is and which questions there are tractable — is the last skill to develop and the one AI currently lacks. AI is, he says, like an extremely technically skilled graduate student: given a sharp, well-posed question, it can do incredibly hard computations correctly, but it does not yet know which question to ask. > *"The difference between a good physicist and a great physicist is knowing what is the right question to ask — that is actually the hardest part of being a scientist."* ## [70:23] Personal evolution from AI skeptic to resident scientist Lupsasca recapitulates his personal arc: skeptic → converted by o3 (which solved in 11 minutes a calculation that would have taken him days) → "AI-pilled" by GPT-5 (which reproduced, in 30 minutes, his best published result on black hole Love numbers and tidal symmetries — a paper whose training cutoff predated its arXiv release) → now resident scientist at OpenAI. He notes that no competing model at the time could match GPT Pro on that calculation. > *"In under 30 minutes, with one hint... it completely solved this problem, which is one of the nicest calculations that I've ever done."* ## [72:46] Solving a black hole perturbation problem with GPT-5 Lupsasca details the "Move 37" moment that converted him: his paper "Why Is There No Love in Black Holes?" establishes new symmetry generators for perturbations of a Kerr black hole (explaining why black hole Love numbers — tidal response coefficients, named after mathematician Augustus Love — are exactly zero). When GPT-5 Pro was first given the full problem cold, it failed. But after being primed with the simpler flat-space warm-up (a 200-year-old known result), it then solved the full Kerr black hole problem in 18 minutes. > *"GPT-5 was able to reproduce one of my hardest calculations, which I think the number of people in the world that could do that you could count on your hands."* ## [76:34] Discussing whether AI can make original, conceptual leaps The hosts ask whether AI is doing genuine recombination versus true creative leaps. Lupsasca cites Terry Tao, who has not yet seen an AI proof that cannot be traced to an obscure reference. But Lupsasca has been impressed and frames the distinction as one of degree rather than kind — humans may also be recombination machines. He believes continued scaling will produce feats of insight that look like creativity, and notes OpenAI is actively working on enabling models to take bigger, more out-of-distribution leaps suited to scientific discovery. > *"I'm not sure there's a qualitative difference. I think it's just a matter of degree — as we continue scaling the capabilities, I don't see why it's going to stop."* ## [80:09] Challenges of 'AI slop' and the future of academic publishing With models now capable of turning out a physics paper in 30 minutes when properly steered, the arXiv preprint server is being flooded with submissions. Lupsasca distinguishes legitimate use (expert steering + careful verification) from "AI slop" — poorly prompted outputs submitted without adequate checking. His proposed response: raise the bar rather than increase volume. The single-minus amplitude papers open a clear line of attack toward genuine quantum gravity questions; the goal should be to pursue harder problems, not to publish incrementally. > *"Instead, I think now that we have this new tool that gives us AI superpowers, I think we should just raise the bar for what it means to write a good paper."* ## [83:13] The bottleneck of writing academic papers Asked what single bottleneck he would remove, Lupsasca nominates the paper-writing process itself — finding it increasingly strange that researchers use AI to do calculations, compress results into a static paper, and then readers feed that paper back into AI to understand it. He envisions interactive, LLM-embedded papers as a plausible future. He also identifies two missing capabilities in current models: (1) the spark of creativity to identify the next important question, and (2) reliable self-verification, so that the onus of checking long AI-generated proofs does not fall entirely on humans. > *"Maybe some kind of interactive paper which lives in some LLM. Maybe your whole paper is some ChatGPT page... I think we're going to head in that direction."* ## [90:19] Final takeaways and looking ahead to the next year Lupsasca's closing message: pay attention. The trajectory from "useful for email" to "solves open problems in quantum gravity" has taken roughly 18 months. Models are solving open problems that expert communities spent years on. Extrapolating forward, with more scaling already in the pipeline, the next 6 to 12 months should bring further surprises. The right posture is excitement, careful verification, and a commitment to pursuing harder problems. > *"If you just extrapolate that into the future, imagine where we're going to be in 6 months or a year — I think it's kind of surreal to live through this time, but it's really happening."* ## Entities - **Alex Lupsasca** (Person): Theoretical physicist, Vanderbilt University professor and OpenAI resident scientist; 2024 New Horizons Breakthrough Prize and IUPAP Young Scientist Award winner; expert in black hole physics and scattering amplitudes. - **Andrew Strominger** (Person): Harvard professor and Lupsasca's former PhD advisor; pioneer of celestial holography; co-author of both single-minus amplitude papers. - **Alfredo Guevara** (Person): Postdoctoral researcher at the Institute for Advanced Study (IAS); performed the foundational hand calculations underpinning the AI-assisted breakthrough. - **David Skinner** (Person): Professor at Cambridge University; co-author of the single-minus gluon amplitude paper. - **Terry Tao** (Person): Fields Medal-winning mathematician at UCLA; referenced regarding the question of whether AI proofs involve genuine creativity. - **Scattering Amplitudes** (Concept): Complex-valued functions in quantum field theory encoding probabilities for particles to scatter; the central mathematical objects of both papers discussed. - **Single-Minus Gluon/Graviton Amplitudes** (Concept): Tree-level scattering amplitudes where all but one particle have positive helicity; previously assumed zero in textbooks but shown non-zero in a collinear phase-space region. - **Parke-Taylor Formula** (Concept): Compact closed-form result for maximally helicity violating (MHV, double-minus) gluon amplitudes derived in the 1980s; the template whose analog was sought for single-minus amplitudes. - **Feynman Diagrams** (Concept): Diagrammatic technique to organize perturbative QFT calculations; individual diagrams represent distinct intermediate-particle histories whose amplitudes are summed. - **Love Numbers** (Concept): Coefficients encoding tidal deformability; famously vanish for black holes, a fact connected to hidden symmetries studied in Lupsasca's "Why Is There No Love in Black Holes?" paper. - **Celestial Holography** (Concept): Research program exploring symmetries of quantum gravity via scattering amplitude structure; motivates studying graviton amplitudes. - **OpenAI** (Organization): AI research company where Lupsasca serves as resident scientist; developer of GPT-5 and the internal extended-reasoning model used for the amplitude proof. - **arXiv** (Organization): Open-access physics and mathematics preprint server; mentioned in the context of AI-generated "slop" flooding submissions. - **GPT-5 / ChatGPT Pro** (Software): OpenAI's frontier language model used as the primary AI tool in both amplitude papers; capable of extended reasoning steps of 20-34 minutes per prompt.