What everyone gets wrong about open source models | First Pass Ep. 8 with Jesse Zhang (Decagon)

Enterprise AI adoption currently favors a transition from frontier models to open-weight models, primarily to optimize latency rather than cost. Jesse Zhang, CEO of Decagon, explains that while frontier models are essential for initial experimentation and understanding use-case requirements, they often contain unnecessary parameters for specific, repetitive production tasks. Once a use case matures, fine-tuning smaller models for targeted jobs—such as topic classification—delivers superior performance and significantly lower latency. The current AI landscape remains in a positive-sum growth phase, where both frontier and open-source models are expanding rapidly. While frontier models capture the majority of early-stage experimentation, the long-term trajectory points toward increased customization via open-source models as enterprises solidify their production workflows and demand greater control over their specific AI infrastructure.