Because pathological tissue slide images enable direct observation of human tissues at the cellular level, pathological diagnosis often serves as the final stage in clinical disease diagnosis. However, pathology images have enormous resolutions, and manual examination of lesion regions by pathologists is not only time-consuming but also heavily relies on their experience and subjective judgment. Furthermore, many quantitative indicators, such as cell count and tumor size, are difficult to calculate accurately manually. Although many pathology AI models have been developed in recent years, most existing pathological AI systems is passively activated by pathologists, and cannot provide the analysis results immediately. In addition, pathological scanners are expensive, and digital pathology images require enormous storage space. It is extremely costly to build a digital pathology system, and most physicians still rely on microscopes for examination and diagnosis. These two issues lead to the lack of fluency in current pathological AI systems, and hinder clinical adoption of pathological AI systems. To address these issues, this project proposes "A Hierarchically Collaborative Autonomous AI Agent Framework for Multimodal Intelligent Pathology Analysis." We propose to utilize the reasoning ability of multimodal large language models to integrate information from pathology images and reports, and build a highly resilient and autonomous AI agent system, endowing it with the ability to autonomously initiate and analyze pathology images. We will also develop model compression techniques to integrate the pathology AI models into microscopes with edge computing. This project aims to reduce hardware construction costs of pathology AI system as well as endow it with autonomous inference capabilities, in order to achieve sustainable development of computational pathology.