Eiso Kant

Eiso Kant is co-founder and co-chief executive of Poolside, which builds foundation models and autonomous agents for software engineering. His central conviction is that executable code—with its testable results and direct connection to real systems—offers a powerful training ground for more capable machine intelligence.

Kant moved from the Netherlands to Spain in 2009 to attend IE University, where he founded recruiting platform Tyba with Jorge Schnura and Philip von Have. The three graduated in business administration in 2013 and built Tyba into an international business connecting young professionals with technology startups, a trajectory Kant recounted in his essay on the company’s origins.

He subsequently built source{d}, applying neural networks to code before modern AI programming assistants became commonplace. Its open-source software projects include gitbase, which makes Git repositories queryable with SQL, and infrastructure for analyzing source code and training machine-learning systems. By 2017, source{d} had developed code-completion technology using LSTMs, attracting acquisition interest from GitHub’s then-chief technology officer, Jason Warner.

Kant declined the acquisition, but the encounter began a lasting partnership. He went on to found and lead Athenian, an engineering-intelligence platform for understanding software-development workflows, and co-hosted Developing Leadership with Warner.

Poolside and the case for executable intelligence

Kant and Warner founded Poolside in 2023. Initially its chief technology officer, Kant now leads the company alongside Warner as co-chief executive. He is also co-founder and president of Poolside Infrastructure Company, which is developing the Project Horizon data-center campus in West Texas.

At AI Engineer Code 2025, Kant operated a Poolside agent that inspected an Ada codebase, translated it into Rust, ran builds and tests, and added command-history navigation. The demonstration illustrated the capabilities he prioritizes: sustained work in unfamiliar systems, execution-based verification, recovery from errors, and tightly controlled permissions.

Poolside subsequently released Laguna XS.2 and Laguna M.1, alongside its pool terminal agent and Shimmer cloud-development environment. Laguna XS.2’s publicly available model weights, released under Apache 2.0, marked a significant expansion beyond the company’s earlier closed enterprise deployments.

Kant’s technical priorities include:

• Reinforcement learning from code execution feedback: Training models against executable outcomes and synthetic examples creates signals beyond next-token prediction while limiting dependence on sensitive customer code, an approach he connects to broader economic productivity.
• Model Factory: Integrating data pipelines, training, evaluation, infrastructure, and deployment makes model development repeatable; Kant has linked this system directly to Laguna’s technical work.
• Minimal agent harnesses: Giving models controlled computing environments and the ability to write and execute scripts supports longer, more adaptable tasks without requiring extensive predefined tool menus.
• Open-weight coding models: Public model releases let external developers inspect, adapt, and extend capable systems, widening participation in software-focused foundation-model development.