Poolside

AI lab building foundation models purpose-trained for software engineering, with a model architecture designed specifically for code generation and understanding rather than general-purpose conversation. Raised $500M+ total; CEO Jason Warner, former GitHub CTO; building custom training infrastructure for code models; Paris and US offices; still pre-revenue but well-capitalized.

Poolside is the most prominent startup building a code-specific foundation model from scratch — if successful, it would be the first model architected for software engineering rather than adapted from general LLMs.

poolside.ai Foundation Models

Franceprivate

At a glance

Total raised

$626M

last round series b

Headcount

51-200

approximate · public range

Headquarters

Paris

France

Founded

2023

3 years active

Status

Private

Foundation Models

Latest valuation

$3B

Series B · 2024

For subscribers

Subscribe to see the full profile

Including Funding rounds, Bull / Bear thesis, Stock + earnings, Roster changes, Patents, News, and Open roles.

Founding

50% off

/month

94 of 100 spots left

Full

$10

/month

Available once all 100 Founding Member spots are claimed.

Get full access

Already subscribed? Sign in →

Funding history

Total raised

$626M

Rounds tracked

Latest stage

Series B

Latest valuation

$3B

Date	Stage	Amount	Valuation	Lead investors
Oct 2, 2024	Series B	$500M	$3B	Bain Capital Ventures
Aug 24, 2023	Series A	$100M	—	Felicis, Xavier Niel
May 21, 2023	Seed	$26M	—	Redpoint Ventures

Founder backgrounds

Eiso Kant

CTO & Co-founder

CTO and co-founder of Poolside. Previously founded source{d} (2015), the first company focused on applying AI to code and software, which open-sourced infrastructure for training code models and was later acquired. Before that, founded Athenian, a data-enabled engineering platform. Earlier ventures include co-founding Tyba and, as a first company, Twollars — a virtual-currency platform for charitable donations. Has been building early-stage companies at the intersection of AI and developer tools for over a decade, with investors often following him across successive ventures.

Jason Warner

CEO & Co-founder

Prior to founding Poolside, spent two years as a managing director at Redpoint Ventures following a tenure as CTO of GitHub, where he oversaw the Office of the CTO and helped scale the platform. Earlier engineering leadership roles include VP of Engineering at Heroku and product engineering for Ubuntu Desktop and Phone at Canonical. Holds degrees from Pennsylvania State University and Rensselaer Polytechnic Institute. Co-founded Poolside with serial entrepreneur Eiso Kant, raising a $26M seed round with a thesis of pursuing "narrow AGI through software and code." Has publicly described himself as a "very average programmer, but an excellent engineer"—a self-assessment that reflects his emphasis on engineering systems and organizational design over individual coding virtuosity.

AI bull / bear

Bull case

Poolside is the only well-capitalized startup building a foundation model architected from the ground up for software engineering, not retrofitted from a general-purpose LLM. That architectural bet matters: a code-native model can optimize for repository-scale context windows, syntax-level reasoning, and deterministic correctness in ways general models cannot. CEO Jason Warner spent years inside GitHub as CTO, giving him unmatched insight into how developers actually work and where existing tools fail. The $626M war chest — including a $500M Series B at a $3B valuation led by Bain Capital Ventures — buys years of runway to train models, build custom infrastructure, and recruit top research talent without revenue pressure. The developer-tools market is enormous, and AI coding assistants are the fastest-growing segment within it. If Poolside ships a model that meaningfully outperforms general-purpose alternatives on real engineering tasks, enterprise adoption could be swift. Distribution through IDE plugins and API endpoints is straightforward; the hard part is model quality, and that is exactly where Poolside is placing its bet.

Bear case

Poolside has raised $626M with zero revenue and no public model. That is an extraordinary amount of capital deployed against an unproven thesis. Meanwhile, general-purpose models from OpenAI, Anthropic, and Google improve on coding benchmarks with every release, and those companies have massive distribution, paying customers, and developer ecosystems already in place. GitHub Copilot alone has over a million paid users and benefits from deep integration into the world's dominant code-hosting platform. Competitors like Cursor and Cognition are shipping product today, not building infrastructure for a model that may arrive years from now. The argument that a code-native architecture will outpace general models assumes the incumbents stand still — a dangerous premise when frontier labs are spending billions annually. At a $3B valuation with no revenue, Poolside must eventually capture a material share of a market where switching costs are low and model quality gaps compress fast. Custom training infrastructure is a capital-intensive gamble that could become a liability if cloud providers offer cheaper, scaled alternatives before Poolside reaches production.

Generated by deepseek-v4-pro · May 2, 2026Refreshes quarterly · see disclosures

In the news

Poolside · last 30 days

No articles ingested yet for Poolside. Once the hourly news pipeline is live, every article the classifier tags as mentioning this company appears here with its one-line AI summary and sentiment.

Leadership

Jason Warner

CEO & Co-founder

in 𝕏

Eiso Kant

CTO & Co-founder

in 𝕏

Margarida Garcia

COO

in 𝕏

Products

Laguna Models
Poolside's family of foundation models purpose-trained from scratch for agentic coding. Laguna M.1 is the flagship model optimized for complex software engineering tasks with a 128K context window, supporting tool calling and multi-step reasoning. Laguna XS.2 is a lightweight, open-weight (Apache 2.0) model built for rapid agentic iteration, using a mixture-of-experts architecture with 33B total / 3B active parameters. Both models are trained on Poolside's own data and infrastructure using reinforcement learning, designed to write code autonomously rather than just offer chat-based suggestions.
pool (Coding Agent)
Poolside's agent runtime that runs coding agents directly from the developer's shell, VS Code editor, or CI/CD pipelines without switching windows. Each agent operates in an isolated sandbox with configurable policy controls, end-to-end audit traces, and version control for agent work. It supports the open ACP standard for agent-to-editor communication and can automate code review, generation, and testing on every push. Agents work inline with the editor, providing suggestions, edits, and context without leaving the development environment.
Shimmer
A web-based, mobile-optimized agentic coding development and interactive preview environment that lets developers write code with Laguna models on the go. Shimmer provides an interactive preview environment for agentic coding workflows, allowing developers to build and iterate on software from any device. It pairs with Poolside's Laguna models to deliver AI-powered coding assistance in a browser-based interface designed for mobility and flexibility.

Open roles

View all careers ↗