AgentOS: From Application Silos to a Natural Language-Driven Data Ecosystem

A team of eight researchers, including Rui Liu and Jian Pei, has published a groundbreaking paper proposing AgentOS, a new operating system paradigm built from the ground up for AI agents. The paper argues that running LLM-based agents like OpenClaw on legacy OSes designed for GUIs and CLIs creates fundamental problems: fragmented interactions, poor permission management (dubbed 'Shadow AI'), and severe context fragmentation. AgentOS proposes a complete architectural shift, replacing the traditional desktop with a unified Natural User Interface (NUI) portal for natural language or voice commands.

At the core of AgentOS is an 'Agent Kernel' that interprets user intent, decomposes complex tasks, and coordinates multiple specialized agents. In this model, traditional applications evolve into modular 'Skills-as-Modules' that users can dynamically compose using natural language rules. Crucially, the researchers frame the realization of AgentOS as a core Knowledge Discovery and Data Mining (KDD) problem. The Agent Kernel must act as a real-time engine for intent mining and knowledge discovery, turning the OS into a continuous data pipeline involving sequential pattern mining for workflows, recommender systems for skill retrieval, and dynamically evolving personal knowledge graphs. This redefinition establishes a new research agenda for the KDD community in building intelligent computing systems.