Listen to the podcast here.

Adaptive CEO and co-founder Julien Launay joined the DeepTechs podcast to discuss the journey from generalist LLMs to specialized AI agents, and why that shift represents one of the most important opportunities in enterprise AI today. Launay traces his path from early work on transformer models and open-source projects like BLOOM, through to co-founding Adaptive in 2023 with a clear mission: solve the last-mile problem of adapting foundation models to the specific needs of large enterprises. The conversation covers how Adaptive's RLOps platform uses post-training and reinforcement learning to build smaller, specialized models that outperform their generalist counterparts on the tasks that matter most.The economics are as compelling as the technical case. Launay explains how enterprises can reduce their AI spend by up to 90% by replacing expensive generalist models with purpose-built specialized agents. For large organizations in telecoms, insurance, and aviation, where AI agents operate at massive scale and generate enormous volumes of data, the cost-performance advantage is transformative. It is a clear and accessible introduction to what Adaptive builds and why it matters.