Jev is to tool use what RAG is to context (rajveerbachkaniwala.com)

TypeSafe has announced Jev, a new model that redefines how output is generated by fixing the available options, much like Retrieval-Augmented Generation (RAG) fixes input. While RAG allows developers to pre-determine the context a model accesses, Jev allows developers to pre-select the tools and choices available for the model to use. This approach streamlines the decision-making process for the AI, increasing speed and predictability while requiring developers to establish the parameters in advance. Jev's architecture represents a significant advancement in AI/ML by creating a more efficient framework for tool selection, which diverges from agentic models that self-determine tool use based on context. By making the model's task narrow and concrete, Jev promises faster and less resource-intensive outputs. The co-existence of Jev and agentic models opens up a dialogue about their comparative effectiveness; for instance, Jev may excel when the choices are known, while agentic models could be more adept in exploratory scenarios. This dynamic reflects an evolving landscape in AI design, where both fixed parameters and adaptable agents play distinct roles in problem-solving.