Artificial intelligence (AI) is evolving from its initial capabilities in perception, recognition, and content generation to more advanced functions such as complex reasoning, task planning, and action execution. The manufacturing industry, after undergoing long-term development encompassing automation, informatization, networking, and platformization, has established a relatively comprehensive foundation in industrial software, equipment connectivity, and data resources. These two trends mutually reinforce each other, offering AI the necessary engineering conditions to penetrate core business processes, including research and development (R&D), production, operations, and services. Consequently, industrial agents have emerged as a pivotal vehicle for AI to empower the new wave of industrialization.

Industrial agents are not intended to replace industrial internet platforms, industrial software, or control systems. Rather, they serve to systematically reintegrate existing digital capabilities to address specific and well-defined industrial tasks. Guided by explicit industrial task objectives, industrial agents gather contextual information relevant to the task, employ specialized models and tools to devise implementation plans, execute these plans in a controlled manner through business systems and on-site equipment, and validate the execution results based on confirmation receipts.

Against this backdrop, the digital and intelligent transformation of the manufacturing industry is experiencing three fundamental shifts: (1) a transition from system functionality-driven to business task-driven approaches, (2) a move from single-point optimization to cross-system collaboration, and (3) a shift from information-assisted decision-making to verifiable and controllable execution. When assessing the development level of industrial agents, it is crucial not to focus solely on the advancement of models but also on their capacity to complete tasks and achieve verifiable business outcomes under real-world conditions. This report presents systematic research findings on the industrial agent sector, organized into five chapters. Chapter 1 defines the concept and strategic significance of industrial agents; Chapter 2 outlines the industrial agent architecture from multiple dimensions, including industry, functionality, and operation; Chapter 3 delves into core technologies across six layers: data, models, networks, computing power, self-evolution, and security; Chapter 4 organizes industrial practices around five stages: R&D and design, pilot testing, production manufacturing, operations management, and marketing services; Chapter 5 envisions the future evolution of industrial agents from auxiliary tools to core engines. Currently, industrial agents are at a critical juncture, transitioning from pilot scenarios to large-scale applications.

This report aims to provide valuable references for refining policy frameworks, advancing the development of application scenarios, and deepening industry-focused research. It also seeks to collaborate with industry stakeholders to explore a development path for intelligent agents that aligns with industrial characteristics, ensures safety and trustworthiness, and supports gradual progress, thereby contributing to the advancement of new industrialization and the establishment of a manufacturing powerhouse. As a new round of technological revolution and industrial transformation unfolds, AI is reshaping production factors, industrial structures, and governance models with unprecedented depth and breadth. AI stands at a critical historical juncture, transitioning from point-based technological empowerment to system-level intelligence. Industrial agents, characterized by autonomous perception, decision-making, execution, and iteration, are becoming the core implementation vehicle for the deep integration of AI and manufacturing. These agents function as task organization and execution systems that bridge model capabilities, software tools, and industrial business scenarios, playing a vital role in driving the large-scale application and industrialization of 'AI + manufacturing.' Their core value lies not in generating more model outputs but in accomplishing industrial tasks such as product design optimization, equipment fault handling, quality anomaly resolution, and order production scheduling. Based on clear information such as task objectives, input conditions, constraints, completion criteria, and responsible entities, industrial agents acquire necessary context, invoke specialized models, industrial software, or equipment capabilities to form strategic plans, verify results based on system feedback or on-site status, and utilize verification conclusions to update subsequent strategies, model parameters, or knowledge bases.

The following are excerpts from the report: