Poolside AI

What Is Poolside AI?

The AI code assistant market has exploded into a multi-billion-dollar category, with GitHub Copilot, Cursor, and a growing roster of competitors all adapting general-purpose language models for software development. Poolside AI is taking a fundamentally different approach: building foundation models from scratch, exclusively for code, trained using execution-based feedback rather than static text prediction.

Founded in early 2023 by Jason Warner (former CTO of GitHub) and Eiso Kant, Poolside operates as a French SAS headquartered in Paris with additional presence in New York, Chicago, Miami, and Madrid. The company raised $500 million in its Series B at a $3 billion valuation in October 2024, with investors including Bain Capital Ventures, Nvidia, eBay Ventures, and DST Global. By late 2025, Poolside was reportedly raising an additional $2 billion at a $12 billion valuation, with Nvidia committing up to $1 billion alone. Revenue reached $50 million in 2025, up from $30 million the prior year.

The company's go-to-market is deliberately narrow. Poolside targets large enterprises with 5,000+ developers — major banks, defence contractors, and government agencies — deploying custom AI models within customer environments rather than offering a generic cloud API. This approach trades market breadth for deep enterprise penetration and premium pricing.

Key Features

Malibu and Point Foundation Models

Poolside's model architecture uses two complementary models. Malibu is the high-capacity model optimised for complex tasks: multi-file code generation, test creation, refactoring, and architectural analysis. It supports context windows exceeding 1 million tokens, enabling it to reason across entire codebases rather than individual files. Point is a smaller, quantised model designed for real-time IDE interactions, delivering sub-200-millisecond code completions. The pairing is strategic — Point handles the low-latency typing assistance while Malibu tackles the complex engineering tasks that require deep contextual understanding.

Reinforcement Learning from Code Execution Feedback (RLCEF)

Poolside's most distinctive technical innovation is its training methodology. Rather than learning solely from static code repositories (the approach most competitors use), RLCEF exposes models to real coding challenges and trains them on execution feedback — actual compiler output, test results, and runtime behaviour. This creates a continuous improvement loop where the model learns not just what code looks like, but whether it works. The approach allows Poolside to generate synthetic training data continuously, breaking through the ceiling of available human-written code for training purposes.

Enterprise On-Premise Deployment

Every Poolside deployment runs within the customer's environment. Through partnerships with AWS (Bedrock and EC2), enterprises can deploy Poolside models inside their existing cloud infrastructure with no code or data leaving their security perimeter. For defence and government clients especially, this architecture is non-negotiable. Models are then fine-tuned on the customer's specific codebase, coding practices, internal libraries, and documentation — creating a bespoke AI coding assistant that understands the organisation's particular technical context.

Continuous Customer-Specific Learning

After initial deployment, Poolside models continue improving from team interactions. As developers accept, modify, or reject suggestions, these signals feed back into the model. Combined with ongoing exposure to the customer's evolving codebase, the AI assistant becomes progressively more attuned to the organisation's coding standards, architectural patterns, and domain-specific requirements. This creates a compounding advantage — the longer a team uses Poolside, the more valuable it becomes.

Pricing

Poolside does not publish pricing. The company operates exclusively through custom enterprise contracts, with no self-serve plan, free tier, or individual developer offering. Given the company's focus on organisations with 5,000+ developers and custom on-premise deployments, contract values likely start in the seven-figure range annually.

The economics are straightforward for Poolside's target market. If an enterprise employs 10,000 developers at an average cost of $150,000 per year, even a 5% productivity improvement from AI-assisted coding represents $75 million in value. Against that backdrop, a multi-million-dollar platform investment is easily justified. The challenge is quantifying that productivity gain with confidence — and Poolside's limited public benchmarking makes independent validation difficult.

For organisations comparing Poolside to GitHub Copilot Enterprise ($39/user/month) or Cursor Business ($40/user/month), the pricing model is entirely different. Those products offer per-seat SaaS pricing for broad adoption. Poolside sells a custom-deployed, continuously-improving AI engineering platform — closer to an enterprise infrastructure investment than a per-user tool subscription.

EU Compliance & Privacy

Poolside AI SAS is a French entity registered in Paris, placing it under EU jurisdiction by default. The company maintains a GDPR-compliant data processing agreement with Standard Contractual Clauses for any data transfers. All enterprise deployments run within customer environments, meaning code and interaction data stay within the customer's infrastructure.

However, there are nuances. Poolside's dual US-France operations mean some corporate functions may operate under US jurisdiction. The company has not publicly disclosed specific security certifications like ISO 27001 or SOC 2 — a gap that larger enterprise buyers will likely push to close. For defence and government clients, Poolside has developed mission-grade security capabilities, though the specifics are not publicly documented.

The on-premise deployment model itself provides strong data sovereignty assurance regardless of the company's corporate structure. Since models run entirely within the customer's AWS, Azure, or private infrastructure, the question of where Poolside processes data becomes moot — the answer is always "within the customer's environment."

Who It's Best For

Large enterprises with 5,000+ developers in financial services, defence, and government who need AI coding assistance without sending proprietary code to external APIs. The on-premise model and continuous fine-tuning on internal codebases deliver value that generic coding assistants cannot match.

Organisations with strict intellectual property requirements where source code is highly sensitive and cannot be processed by third-party cloud services under any circumstances.

Teams seeking code-native AI rather than repurposed general-purpose LLMs. Poolside's RLCEF training methodology and code-specific architecture provide an advantage for complex software engineering tasks.

Not suitable for individual developers, small teams, or startups. The enterprise-only model, absence of self-serve access, and high contract values place Poolside firmly outside the reach of any organisation that does not have thousands of developers and an enterprise procurement process.

The Verdict

Poolside AI represents the most ambitious bet in the AI code assistant space: custom foundation models trained specifically for code, deployed within enterprise environments, and continuously improved from team interactions. The technical approach — RLCEF training, code-native architecture, million-token context windows — addresses genuine limitations of competitors that adapt general-purpose LLMs for coding tasks. Jason Warner's experience as GitHub CTO gives the founding team credibility that few competitors can match. The main concerns are accessibility and maturity. With no public product, no self-serve access, and limited transparent benchmarking, evaluating Poolside requires a bespoke enterprise engagement. The $12 billion valuation at this stage reflects investor confidence in the vision, but the product still needs to prove its superiority over increasingly capable alternatives from GitHub, Cursor, and others in open competition.

Frequently Asked Questions

Is Poolside AI GDPR compliant?

Yes. Poolside AI SAS is a French company registered in Paris. All enterprise deployments run within customer environments — no code or data leaves the customer's infrastructure. The company maintains a GDPR-compliant data processing agreement with Standard Contractual Clauses.

Who founded Poolside AI?

Poolside was founded in 2023 by Jason Warner, former CTO of GitHub, and Eiso Kant, a software entrepreneur. The team brings deep experience in developer tooling, having helped scale GitHub's platform to over 100 million developers.

How is Poolside different from GitHub Copilot?

Poolside builds its own foundation models exclusively for code using RLCEF training, deploys within customer environments (no data leaves your infrastructure), and fine-tunes on your specific codebase. GitHub Copilot uses adapted general-purpose models via a cloud API. Poolside targets enterprises with 5,000+ developers, while Copilot serves individual developers through organisations of all sizes.

Can individual developers use Poolside?

Not currently. Poolside focuses exclusively on large enterprise deployments with custom contracts. There is no self-serve plan, free tier, or individual developer access. The company's go-to-market targets organisations with 5,000+ developers.

What is RLCEF?

Reinforcement Learning from Code Execution Feedback is Poolside's proprietary training approach. Unlike traditional training on static code, RLCEF trains models using actual execution results — compiler output, test results, and runtime behaviour. This produces models that understand whether generated code works, not just what it looks like.