--- title: Inside the Model Factory — Eiso Kant, Poolside AI description: Poolside AI co-CEO Eiso Kant discusses the company's journey from spending $12M on code language models before the market cared, to building a 'Model Factory'... canonical: https://daily.dev/posts/inside-the-model-factory-eiso-kant-poolside-ai-impk4usz3 twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: Inside the Model Factory — Eiso Kant, Poolside AI | daily.dev og:description: Poolside AI co-CEO Eiso Kant discusses the company's journey from spending $12M on code language models before the market cared, to building a 'Model Factory'... og:url: https://daily.dev/posts/inside-the-model-factory-eiso-kant-poolside-ai-impk4usz3 og:image: https://api.daily.dev/og/posts/IMpK4USZ3.png og:image:alt: Inside the Model Factory — Eiso Kant, Poolside AI og:image:width: 1200 og:image:height: 630 og:locale: en --- > ## Documentation Index > Fetch the complete documentation index at: https://daily.dev/llms.txt > Use this file to discover all available pages before exploring further. # Inside the Model Factory — Eiso Kant, Poolside AI **[Latent Space](https://daily.dev/sources/latentspace)** · 123 min read · 0 upvotes · 0 comments ## Summary Poolside AI co-CEO Eiso Kant discusses the company's journey from spending $12M on code language models before the market cared, to building a 'Model Factory' capable of taking a model from pre-training to release in five to eight weeks. The conversation covers Poolside's engineering-first philosophy (90% engineering, 10% research), how fewer than 70 researchers run 10,000–20,000 experiments per month using streaming data, immutable data layers, and reproducible experimentation. Eiso explains the Laguna S model (118B total, 8B active parameters), noting its gains come from behavioral improvements like persistence, verification, and backtracking rather than raw intelligence. Topics also include why Poolside embraced open weights and open research, their view that MCP and traditional tool calls are flawed, the role of agents inside the model factory, RL moving earlier into pre-training, and why Eiso would rather see 100 foundation model companies than five. ## Full article daily.dev links to this article rather than hosting it. Read it at the original source: ## Questions this post answers ### What is Poolside's Laguna S model and how big is it? Laguna S is a sparse mixture-of-experts language model with 118 billion total parameters and 8 billion active parameters, released as Laguna S 2.1. It was trained in FP8 precision, took eight weeks from the start of training to launch, and outperforms Thinking Machines' roughly 1 trillion parameter open-weight model on several benchmarks despite being far smaller. _Developers evaluating small versus large coding models can track releases like this on daily.dev._ ### How does Poolside train and release new models so quickly? Poolside built an internal 'Model Factory' that streams training data directly into jobs instead of pre-packaging datasets, treats data as immutable and code as versioned for full reproducibility, and increasingly uses coding agents to write pipeline code, launch training jobs, and evaluate results. With fewer than 70 researchers, the team runs 10,000-20,000 experiments monthly and cut model-release cycles from six months down to five and eight weeks. _Engineers curious about ML infrastructure practices can follow model-factory approaches like this via daily.dev._ ### Why does Poolside think MCP and traditional tool calls are a bad design for AI agents? Poolside's co-founder argues that stuffing dozens of predefined tools into a system prompt is inefficient compared to giving a model a minimal harness, a container, and a code base where it can write scripts with conditionals and loops to interact with data sources directly. Their Laguna S model already shows this shift, and they predict system prompts crammed with 20-40 tools will largely disappear within about a year. _Teams designing agent harnesses can weigh tool-calling versus code-execution approaches using coverage like this on daily.dev._ ## Similar posts on daily.dev - [Laguna XS.2 and M.1: A Deeper Dive — Poolside](https://daily.dev/posts/laguna-xs-2-and-m-1-a-deeper-dive-poolside-ko2qlww4d) · Hacker News · 0 upvotes · 0 comments - [Laguna by Poolside: Foundation models for agentic coding and long-horizon work](https://daily.dev/posts/laguna-by-poolside-foundation-models-for-agentic-coding-and-long-horizon-work-p551ccehq) · Product Hunt · 2 upvotes · 0 comments --- Tags: [#machine-learning](https://daily.dev/tags/machine-learning), [#llm](https://daily.dev/tags/llm), [#ai-agents](https://daily.dev/tags/ai-agents), [#reinforcement-learning](https://daily.dev/tags/reinforcement-learning) [View this post on daily.dev](https://daily.dev/posts/inside-the-model-factory-eiso-kant-poolside-ai-impk4usz3) ```json {"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://daily.dev/#organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180},"sameAs":["https://twitter.com/dailydotdev","https://github.com/dailydotdev","https://www.linkedin.com/company/daily-dev-ltd"]},{"@type":"WebSite","@id":"https://daily.dev/#website","url":"https://daily.dev","name":"daily.dev","publisher":{"@id":"https://daily.dev/#organization"},"potentialAction":{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https://daily.dev/search?q={search_term_string}"},"query-input":"required name=search_term_string"}}]} {"@context":"https://schema.org","@type":"TechArticle","headline":"Inside the Model Factory — Eiso Kant, Poolside AI","url":"https://daily.dev/posts/inside-the-model-factory-eiso-kant-poolside-ai-impk4usz3","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/inside-the-model-factory-eiso-kant-poolside-ai-impk4usz3"},"datePublished":"2026-07-23T05:11:59.602Z","dateModified":"2026-09-13T21:19:23.908Z","description":"Poolside AI co-CEO Eiso Kant discusses the company's journey from spending $12M on code language models before the market cared, to building a 'Model Factory'...","image":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/66b7c30004e0b80b8a60571d93436995?_a=AQAEuop","thumbnailUrl":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/66b7c30004e0b80b8a60571d93436995?_a=AQAEuop","isAccessibleForFree":true,"articleSection":"Latent Space","inLanguage":"en","publisher":{"@type":"Organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180}},"author":{"@type":"Organization","name":"Latent Space","logo":"https://media.daily.dev/image/upload/s--DnJ9laFj--/f_auto,q_auto/v1780213196/logos/latentspace?_a=BAMAMiWQ0","url":"https://daily.dev/sources/latentspace"},"commentCount":0,"discussionUrl":"https://daily.dev/posts/inside-the-model-factory-eiso-kant-poolside-ai-impk4usz3","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":0},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":0}],"keywords":"machine-learning,llm,ai-agents,reinforcement-learning","timeRequired":"PT123M"} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"Latent Space","item":"https://daily.dev/sources/latentspace"},{"@type":"ListItem","position":3,"name":"Inside the Model Factory — Eiso Kant, Poolside AI"}]} {"@context":"https://schema.org","@type":"FAQPage","@id":"https://daily.dev/posts/inside-the-model-factory-eiso-kant-poolside-ai-impk4usz3#faq","mainEntity":[{"@type":"Question","name":"What is Poolside's Laguna S model and how big is it?","acceptedAnswer":{"@type":"Answer","text":"Laguna S is a sparse mixture-of-experts language model with 118 billion total parameters and 8 billion active parameters, released as Laguna S 2.1. It was trained in FP8 precision, took eight weeks from the start of training to launch, and outperforms Thinking Machines' roughly 1 trillion parameter open-weight model on several benchmarks despite being far smaller. Developers evaluating small versus large coding models can track releases like this on daily.dev."}},{"@type":"Question","name":"How does Poolside train and release new models so quickly?","acceptedAnswer":{"@type":"Answer","text":"Poolside built an internal 'Model Factory' that streams training data directly into jobs instead of pre-packaging datasets, treats data as immutable and code as versioned for full reproducibility, and increasingly uses coding agents to write pipeline code, launch training jobs, and evaluate results. With fewer than 70 researchers, the team runs 10,000-20,000 experiments monthly and cut model-release cycles from six months down to five and eight weeks. Engineers curious about ML infrastructure practices can follow model-factory approaches like this via daily.dev."}},{"@type":"Question","name":"Why does Poolside think MCP and traditional tool calls are a bad design for AI agents?","acceptedAnswer":{"@type":"Answer","text":"Poolside's co-founder argues that stuffing dozens of predefined tools into a system prompt is inefficient compared to giving a model a minimal harness, a container, and a code base where it can write scripts with conditionals and loops to interact with data sources directly. Their Laguna S model already shows this shift, and they predict system prompts crammed with 20-40 tools will largely disappear within about a year. Teams designing agent harnesses can weigh tool-calling versus code-execution approaches using coverage like this on daily.dev."}}]} ```