Knowledge Base For Startups

source_url: https://aws.amazon.com/startups/learn/hpoolside-pioneers-ai-assisted-software-development-on-awsow title: "How poolside pioneers AI assisted software development on AWS"

How poolside pioneers AI assisted software development on AWS

See how our partners at poolside are turning millions of developers into billions

Modern human history is written in code, and much of how we live and work is defined by software. Today, there are approximately 100 million developers in the world, each contributing to a society increasingly shaped by technology. But there are only so many hours in the day, and developers are all limited by the human capacity to architect, code, debug, test, refactor, learn, collaborate—the list goes on. At least, that's how it used to be. As artificial intelligence (AI) evolves at an unparalleled rate, that paradigm is changing fast.

Next generation AI company poolside is pioneering the world's most advanced AI for software development: an AI that may one day impact every new line of code written in the world. Enterprise leaders can empower their developers with poolside's industry-leading AI capabilities, from coding companions to automated code completion—all powered by cutting edge foundation models (FMs) built for software.

Together with Amazon Web Services (AWS), poolside is expanding and simplifying access to its proprietary models in Amazon Bedrock, and powering cost-efficient inferencing on AWS Trainium. Eventually, the company hopes to turn millions of developers into billions—each empowered to create software that propels humanity forward.

The road to artificial general intelligence (AGI)

Image: poolside founders, Jason Warner (CEO) and Eiso Kant (CTO) sitting outside

Founded in 2023 by Jason Warner (CEO) and Eiso Kant (CTO), poolside is rooted in the shared belief that AGI is not only possible, but probable. "In an AI-driven world, we realized that the next 18-36 months in front of us will see the distance between human capabilities and machine intelligence close dramatically," says Kant. "We started poolside because we have a unique view on how to close that gap faster." Warner adds: "If you have that belief, it's the only thing you're going to work on because from where you're standing, the arc of humanity bends in that direction."

For poolside, the pathway to AGI runs through software and the empowerment of developers. "We're going to live in a developer-led world that is becoming increasingly AI assisted. poolside is the only company building foundation models and a product experience that are entirely oriented on AI for software development," says Kant. Warner adds: "We have the potential to impact every single line of code, every developer, and eventually every person on the planet. We exist to serve them on our way to AGI."

"When humans have access to intelligence bound only by compute, the time needed to solve complex problems is significantly reduced," says Warner. "Suddenly, projects that took 10 developers six months to complete can be done in a matter of compute hours." And that's only the beginning—poolside is working towards a future where the number of developers on the planet approaches that of the global population itself. "Imagine a world where over 8 billion people have that superpower," says Warner.

Unlike other AI companies, poolside provides a full stack solution. This means it trains its own models and builds applications like the poolside generative AI Assistant. Today, the company is built around two proprietary large language models (LLMs), point and malibu. Kant explains: "point offers us the ability to place code completion capabilities in the hands of developers. malibu is our flagship—and largest ever—model, built for the most challenging software engineering problems like code generation, testing, refactoring, and more."

AI models that learn like humans

The techniques used to train poolside's models are a key differentiator for the company, specifically, Reinforcement Learning from Code Execution Feedback (RLCEF). RLCEF allows poolside to break the ceiling of code data available by generating synthetic code continuously, and training its models on increasingly more data every day without having to rely on its customers' interactions with the model.

"Our models aren't just learning by consuming code; they're learning by coding themselves," says Kant. "We take our model and put it in an environment of hundreds of thousands of code bases, and they essentially go and explore solutions to millions—soon to be billions—of coding tasks. They learn from when they're right and when they're wrong, just like we do ourselves."

Infrastructure fit for the AI frontier

Any AI company needs access to scalable infrastructure and significant computing resources. "The sheer scale of compute required for training increasingly capable foundation models is key to our business," says Kant. "When we started, we knew that within 12 months we would need at least 10,000 GPUs to remain in the AI race." Having spent time with multiple providers, the poolside team decided that AWS aligned with their immediate and long-term technical needs. For instance, the company plans to use AWS Trainium2 chips for model training next year.

Trainium2 chips are purpose built for deep learning (DL) and generative AI workloads. They are optimized to train and deploy the most demanding models including LLMs, multimodal models, diffusion transformers, and more to build a broad set of generative AI applications. Trainium2 is expected to deliver savings of 40 percent for poolside, and additional value to its customers who can reduce their own inference costs by running poolside models on Trainium chips.

The perfect culture match

One of poolside's greatest strengths is the deep expertise of its team. Prior to founding the business, Warner worked as CTO at GitHub, while Kant built his first venture, source{d}, (the world's first company to dedicate itself to applying AI to source code and software). Together, they are building a team that combines technical aptitude and a tireless work ethic.

When poolside sought a cloud service provider, the team needed one that could match its intensity and focus—and AWS delivered. "The main reason we chose AWS was culture," says Kant. "We are a massively execution-driven organization, and the AWS culture of customer obsession and extremely hard work ethic align perfectly with that."

"When you're building your own company, it's not the same as if you were inside another entity building a product," says Warner. "Partnering with AWS, to us, felt like partnering with somebody who understood that this is our life's work. That was the most important bit for me. If I'm betting my company on somebody, I need to feel that those sitting on the other side of the table understand the gravity of that."

poolside arrives in Bedrock

Beyond the technical and day-to-day working aspects of the partnership, AWS is also supporting poolside's go-to-market efforts. To make its offerings as accessible—and affordable—as possible, poolside's malibu and point models were recently made available in Bedrock. "We're a first party offering on AWS, meaning customers can access poolside like they would any AWS service. That's incredibly valuable for customers because they now have access to our advanced AI on some of the world's best infrastructure." says Warner.

Bedrock is a fully managed service that delivers access to high performing FMs through a single API. This enables businesses to easily evaluate, experiment with, and privately customize FMs. Bedrock is serverless, alleviating the need for technical teams to manage infrastructure, and is equipped with deep security features spanning everything from full data encryption to robust access management capabilities.

Availability in Bedrock, alongside Trainium, enables poolside customers to reduce costs. "GPUs are expensive and the AI accelerators that AWS has built allow us to deliver the poolside experience at a highly reduced cost. Ultimately, our customers care about experiencing the product and the intelligence of the underlying models. They're not thinking about the silicon," says Eiso.

Becoming a first party offering on AWS also helps poolside to gain exposure to a wider customer base. "Our go-to-market is combined," says Warner. "AWS has earned the right to be in rooms across the world. Those rooms also matter to us and now, wherever they go to market, it's possible for us to go to market as well."

A race to the moon

Going forward, poolside and AWS are continuing to innovate on the AI frontier. "We joke that we're married to AWS, but it really does feel like that. I spend as much time speaking with AWS people as I do with our own. It's become one team, and that's something that I have never experienced with a cloud provider before," says Kant.

"We have to build a company that serves developers in every corner of the planet," says Warner. "2025 is about advancing our own state of the art capabilities and getting those capabilities into the hands of as many developers on the planet as possible."

"We're on the path to AGI and, for the first time in my life, I feel like there's an event horizon that I can't see past. This is bigger than mobile, bigger than the internet, bigger than personal computers. This is more akin to electricity, and the third order effect of electricity was putting people on the moon."

Image: Jason Warner (CEO) and Eiso Kant (CTO) laughing together

Source: AWS Startups