Universal Deep Research: Bring Your Own Model and Strategy

The paper addresses the emerging paradigm of AI-powered agentic systems, specifically Deep Research Tools (DRTs) that leverage Large Language Models for automated information synthesis. The dominant paradigm involves DRTs with fixed, developer-defined research strategies. This work challenges that model by identifying a crisis of rigidity and lack of user control. It proposes a new paradigm where the research strategy is decoupled from the agent itself, becoming a dynamic, user-specified input that directs the system's behavior, fundamentally changing the architecture and locus of control in these systems.

This paper introduces Universal Deep Research (UDR), a system that moves beyond refining existing research agents and instead proposes a fundamentally new architecture. It identifies a crisis in the current paradigm, where hard-coded strategies limit user control and specialization. The proposed revolutionary model allows users to define research strategies in natural language, which are then compiled into executable code. This decouples the "how" from the "what," shifting agentic systems from static tools to dynamic, user-programmable platforms. This conceptual shift from a fixed to a programmable agent framework constitutes a model revolution in the design of research tools.