Inside the Model Factory — Eiso Kant, Poolside AI
July 23, 2026
AI Summary
5 min readEiso Kant, co-founder of Poolside AI, started his first AI company in 2015 after reading Andrej Karpathy's "The Unreasonable Effectiveness of Recurrent Neural Networks." That company, Sourced, spent four years building language models on code and burned $12 million before failing. "It was the biggest failure of my career," Kant says. He didn't look at language models again for two years. When ChatGPT launched, former colleagues texted him old decks and talks — vindication, but cold. When he started Poolside three years ago, he made two bets that were not obvious at the time: that LLM capabilities would keep compounding, and that reinforcement learning would be the biggest driver of progress. The company never thought about open source again until early this year, when Kant and his co-founder Jason realized the world was heading toward an oligopoly of intelligence — a future that looked like a dystopian sci-fi novel to a self-described utopian. So they pivoted back to openness, not because it was easy, but because they'd rather live in a world with a hundred foundation model companies than five, even if they were one of the five.
The Model Factory: Engineering as the Differentiator
Continue reading the full summary in the app — free to try.
Read Full Summary →Free • No credit card required
Never miss an episode of Latent Space: The AI Engineer Podcast
Get every new episode summarized in your inbox — free, ~5 minutes to read.
No spam. Unsubscribe anytime.
What you'll learn
- 1 (00:58) **From Karpathy to Poolside: The Origin Story** - Eiso explains how Andrej Karpathy's 2015 blog post on RNNs led him to pivot his startup, Sourced, to build language models on code years before the field existed.
- 2 (04:04) **Why Poolside Went Open Source (And Why It Almost Didn't)** - The decision to open-source was not the original plan; the company started with a closed, AGI-first mission.
- 3 (08:37) **The Western Open-Source Gap & Global Talent Strategy** - Poolside is an American company that deliberately avoided hiring researchers in the Bay Area to win the global talent war.
- 4 (13:58) **The Adam Optimizer Bug That Built Their Intuition** - A three-week debugging saga over the epsilon parameter in Adam taught them to trust their own intuitions over papers.
- 5 (16:04) **The Model Factory: An Industrialized Process for Training** - Poolside treats model building as 90% engineering, building a factory from raw data to deployment, not just a single training run.
- 6 (19:27) **The Factory's Output: 5-Week Model Cycles** - Laguna Access 2 went from pre-training start to launch in 5 weeks; the new model took 8 weeks.
- 7 (20:37) **Agents Inside the Factory: Early Signs of RSI** - The model factory's engineering APIs are perfect for agents, which are now writing code, launching jobs, and evaluating results for researchers.
+ Full timestamped outline available in the app
Show Notes
In recent months, the open vs closed, and US vs China discussions on model ownership and sovereign/local AI have heated up to a fever pitch. So it is very very good news that Poolside AI are finally emerging with new models, like Laguna S 2.1, that are beating Thinking Machines’ recent release nearly 10 times their size.
Poolside’s recent tech report got a lot of praise due to their level of detail, and Vibhu first covered Laguna’s recent technical report on our paper club:
From spending $12 million building language models for code before the world cared to creating a Model Factory that can take a model from pre-training to release in eight weeks, Eiso Kant has spent more than a decade betting that code is the path to AGI. In this episode, the Poolside co-founder joins swyx and Vibhu to explain why ChatGPT felt like vindication, why Poolside embraced open weights and open research, and why he would rather live in a world with 100 foundation model companies than five even if Poolside were one of the five.
We go deep on Poolside’s Model Factory: the engineering systems behind 10,000–20,000 experiments per month, streaming data directly into training, reproducible experimentation, low-precision compute, and agents that increasingly write code, launch jobs, evaluate results, and modify the pipelines used to train future models. Eiso also unpacks their recent launch Laguna S, why persistence, verification, and backtracking may matter more than raw intelligence, how much capability remains inside smaller models, why reinforcement learning will move earlier into pre-training, and why next-token prediction is still extracting too little from the web.
We also discuss model-harness co-design, Poolside’s path from coding agents to AGI, why Eiso thinks MCP and traditional tool calls are “stupid,” the real economics behind frontier-model training, Poolside’s $500 million raise, open-source AI, regulation, NVIDIA and TSMC’s influence, engineering productivity in the agent era, high-agency teams, and hiring at Poolside.
We discuss:
* How Andrej Karpathy’s RNN work inspired Eiso to start building language models for code in 2015
* Why Eiso spent four years and $12 m
More from this podcast
Latent Space: The AI Engineer Podcast →