Inside the Model Factory — Eiso Kant, Poolside AI

Poolside founder Eiso Kant details the company's mission to democratize artificial intelligence through open-source foundation models. The discussion centers on the "model factory" approach, which treats model building as an industrialized, engineering-heavy process rather than purely experimental research. By emphasizing reproducibility, immutable data layers, and streaming data into training, the team achieves rapid development cycles, exemplified by the five-week creation of the Laguna S model. Kant argues that persistence and behavioral refinement—rather than just parameter scaling—are critical for achieving high-performance results in coding and long-horizon tasks. The conversation also addresses the strategic necessity of maintaining a diverse ecosystem of foundation model providers to prevent an intelligence oligopoly, while acknowledging the ongoing need for responsible safety practices as models approach the threshold of recursive self-improvement.