Here’s what’s on our plate today:

• 🍳 Poolside’s AGI bet, code-first training, and investor confidence.

• 📰 Discord age checks, India’s AI land grab, and ir-taxi patents.

• 🧪 Three weekend tools to test agents, coding copilots, and infra.

• 📊 Today’s poll on where institutional AI money really flows.

Let’s dive in. No floaties needed…

The Laboratory

Why big-ticket investors are betting on Poolside

An important factor that helps companies stand out in a crowded market is their ability to attract investor interest in the early stages of development. At the most basic level, investment signals that venture capital firms, strategic investors, and institutional funds are betting that the company has the potential to become extremely valuable. Continued investor trust also signals that the company has moved beyond ideation and has an effective roadmap to deliver value.

Additionally, because other market participants, including customers, employees, competitors, and the media, view investment as a credibility signal, a heavily funded startup is assumed to have undergone some level of technical and commercial scrutiny.

In today’s market, with investors already allocating significant capital to artificial intelligence, standing out requires more than a solid business plan. It requires a long-term vision for AI’s future and a plan to bring it to life.

Why investors care

In such conditions, one company has grown from a 2023 startup to a well-known, well-funded company valued at $12B as of October 2025.

Poolside’s rapid ascent stems not only from its promise of growth in the crowded AI space but also from the perception that it is a driver of artificial general intelligence.

Before Poolside existed

The story of Poolside began way before the company was officially launched. The proverbial tilling of the ground began in 2017, when Jason Warner, then running engineering at GitHub, sought to acquire a small startup called source{d} that applied machine learning to source code.

The deal never closed, but Warner and Eiso Kant, founder of source{d}, became friends. Over the next six years, as Warner oversaw the creation of GitHub Copilot and watched it grow into the most widely used AI developer tool worldwide, he became convinced of two things. First, AI-generated code will eventually become the default approach to building software. Second, that the general-purpose models powering tools like Copilot would never be sufficient for the enterprises that needed them most: banks, defense contractors, and government agencies.

This understanding led Warner to leave his venture capital role at Redpoint in 2023 to co-found Poolside with Kant.

Rethinking how models learn code

The philosophy behind Poolside was to build a company that would stand out not for its product interface or its sales pitch, but for how it trains its models.

To achieve this, Poolside developed a method in which models not only generate code by reading large volumes of existing code scraped from the internet, but also write, execute, and learn from the results.

The method, called Reinforcement Learning from Code Execution Feedback, or RLCEF, mirrors how a human developer improves. The idea is that developers do not improve by reading other people’s code, but by writing their own, running it, seeing it fail, figuring out why, and trying again.

The simple idea propelled Poolside to build the world’s largest code execution environment: its models explore solutions for millions of tasks across over 130,000 real-world code repositories, receiving execution feedback for every attempt. The system powers roughly 1M container images and handles 10,000 code executions per minute.

CTO Kant has stated that 20% of Poolside’s training data is now synthetically generated, and that this proportion will approach 98% in the coming years. “We don’t believe in spending billions on human-labeled data to polish models,” Kant said at a 2025 conference.

The AGI bet behind the strategy

The implications of Poolside’s approach extend well beyond managing investor trust and building strategic partnerships with names such as AWS, CoreWeave, NVIDIA, eBay, and many more.

The real implication is that if RLCEF works at scale, Poolside could generate an essentially unlimited supply of high-quality training data without relying on either the open internet or its customers’ proprietary code.

Simply put, Poolside is opening a new path for AGI.

Poolside hypothesizes that code is the fastest route to general intelligence because it is the only domain where AI can receive deterministic, verifiable feedback on its outputs. You can run code and know instantly whether it works. You cannot do that with most other knowledge work.

Their method, RLCEF, trains models by having them write, execute, and learn from code; as such, the argument is that this process builds not just coding ability but the deeper capabilities that define AGI: multi-step reasoning, long-horizon planning, and the ability to understand why something works, not just what it looks like.

Additionally, the synthetic data flywheel removes the primary constraint that slows the development of more powerful AI models: data. Each code execution generates new training data, so Poolside can theoretically train on an unlimited supply of high-quality tokens without hitting a data wall.

Enterprise as a revenue engine

Even if the company’s AGI approach fails to meet its stated goals, it has also invested in building data centers and tools for enterprises in the defense, government, and financial sectors.

The company builds custom AI models for each customer, trained on that organization’s own codebase, coding standards, and development history, all running entirely within the client’s security boundary. No data ever leaves the perimeter. It integrates with standard tools such as VS Code and JetBrains and is available via Amazon Bedrock for managed deployment in private cloud environments.

While the company continues to pursue the AGI goal, revenue continues to flow in through enterprise contracts that bundle models with professional services, rather than seat-based licensing.

Owning the compute layer

The founding team’s ambitions extend beyond software. Poolside is not just building models; it is also building the power plant behind them. Its Project Horizon is a 2-gigawatt AI campus in West Texas, set to house 40,000+ Nvidia GPUs and become one of the largest data centers in the US.

CoreWeave is the anchor tenant under a 15-year lease. The goal is vertical integration to ensure that the company controls energy, computing, and infrastructure rather than renting them. For a two-year-old startup, such an infrastructure commitment is extraordinary and signals ambitions well beyond software.

What does this mean for the AI competition?

In a crowded AI market, Poolside has attracted significant investment and built a solid foundation for continued scaling.

The company presently faces competition from Copilot’s 77,000 enterprise customers, Cursor’s $300M+ ARR, and Meta’s Code Llama and DeepSeek-Coder, which offer free self-hosted alternatives. The silver lining is that Gartner predicts 75% of enterprise software engineers will use AI code assistants by 2028.

In this environment, the company has articulated the idea that the future of AI is not one model to rule them all, but a constellation of specialized systems trained to perform specific tasks better than anything general-purpose ever could.

Poolside then has investors’ attention and backing for reasons that go far beyond what simple financial statements reveal. The company is still in its early stages of impacting the wider industry. However, if one looks at the speed of its growth and the vision driving it, the logic behind big-ticket investors backing Poolside becomes clear.

Quick Bits, No Fluff

• Discord’s ID dragnet: Researchers say Discord’s new age checks lean on Persona’s broader data network, raising fresh privacy and surveillance concerns.

• India’s AI land grab: Indian platforms are willingly burning near-term revenue as they race to lock in users and dominate the country’s AI boom.

• Air taxi patent dogfight: Archer and Vertical are trading lawsuits over eVTOL designs, threatening to slow commercial air taxi launches before they really start.

Thursday Poll

3 Things Worth Trying

• Continue.dev: VS Code and JetBrains Copilot that watches how you work, then helps refactor, autocomplete, and run multi-step edits locally.

• Sourcegraph Cody: AI code assistant that understands your entire monorepo, ‘answers where is this used’ questions, and drafts changes across many files.

• Plandex: Agentic dev tool for long-running coding tasks that need planning, context, and iterative execution rather than one-off completions.

Rate This Edition