Click the blue text to follow us
Agents Transition from Concept to Core of Industry
On July 28, the 2025 World Artificial Intelligence Conference (WAIC) concluded in Shanghai. This year’s conference, themed “Intelligent Era, Together in Harmony”, attracted over 1,200 top experts and more than 800 companies, showcasing over 100 “global debut” AI products, marking the largest scale in history. Among them, AI Agents became one of the most focused topics. From OpenAI’s “Autonomous Task Execution Agent” to Huawei’s industry-level intelligent workflow, from startups breaking through in vertical scenarios to tech giants’ ecological layouts, Agent technology is moving from the laboratory to large-scale commercial use, becoming a key engine driving AI implementation.
1. Highlights of Agents at WAIC 2025:
From “Tools” to “Colleagues”
At the 2025 World Artificial Intelligence Conference, AI Agents became the focal point, with numerous cutting-edge products and innovative ideas passionately colliding.
OpenAI’s ChatGPT agent made a stunning appearance, showcasing powerful capabilities that seem to open a new door to future intelligent work. It integrates the ability to interact with operators and websites, along with Deep Research’s information integration skills, combined with ChatGPT’s intelligent dialogue advantages, allowing it to easily call software like Excel and CRM systems, deeply analyze sales data, meticulously write reports, and accurately send emails with over 90% accuracy. Moreover, it can autonomously create beautiful PPTs, book hotels, and even plan the best itinerary for visiting over 30 Major League Baseball stadiums in the U.S., generating intuitive visualized Excel spreadsheets. OpenAI’s CTO Mira Murati revealed that the next generation of ChatGPT will further deeply integrate multimodal Agent capabilities, aiming to achieve “one-sentence requests, full-process execution”, a grand goal that will undoubtedly once again overturn people’s imagination of artificial intelligence.
Huawei’s Pangu Agent 3.0 also shone at the conference, focusing on finance and manufacturing, becoming a powerful tool for promoting industry intelligent transformation with its launched “Agent Workflow Engine”. In industrial applications, it can optimize production processes and significantly improve equipment failure prediction accuracy; in finance, it can assist in risk assessment and decision-making; in medical scenarios, it can aid in medical image analysis by combining visual, voice, and text understanding. Huawei’s rotating chairman Hu Houkun asserted that in the next five years, Agents will take over 40% of enterprise procedural work. This confident prediction is backed by Pangu Agent 3.0’s profound understanding and precise grasp of industry business processes, leveraging the construction and application of industry knowledge graphs to clearly outline the interaction processes between different systems, achieving efficient automated collaboration across systems through the autonomous decision-making and execution capabilities of AI Agents, fully demonstrating the immense potential of AI Agents to deeply integrate with existing business systems and drive digital transformation in the industry.
Startups are also actively exploring vertical scenarios, becoming a new force in the application of AI Agents. The “Intelligent Cockpit Agent OS” jointly launched by Jietian Xingchen and Geely Automobile attracted much attention at the conference, acting as a thoughtful and intelligent in-car partner, achieving multimodal natural interaction. By keenly recognizing user voice, gesture commands, and deeply mining user historical behavior data, it constructs accurate user profiles and behavior models, proactively recommending personalized services to greatly enhance the user’s cockpit experience. The advertising industry Agent showcased by MiniMax, based on precise insights into the advertising industry’s business processes and market rules, cleverly integrates natural language processing and image generation technologies to autonomously complete the entire process from market analysis, creative design to media placement, saving 70% of labor costs and building a complete business loop, validating the feasibility of integrating AI technology in vertical industries and enhancing efficiency through Agents.
2. Three Major Trends of AI Agents in 2025
From “Single Task” to “Full Process Autonomy”
Standardization of Industry-level Agents
Upgrading Human-Machine Collaboration Models
From “Single Task” to “Full Process Autonomy”
Early Agents could only execute simple commands (like checking the weather), but now they can link multiple tasks (for example, “planning a marketing campaign” requires coordinating design, budget, placement, etc.). The logistics Agent “Hymala” from Xijing Technology can autonomously schedule port container transport, reducing manual intervention. From a theoretical evolution perspective, this trend reflects the transition of AI Agents from simple task execution units to complex intelligent decision-making systems. Early single-task Agents primarily demonstrated intelligence through accurate execution of single commands, based on simple rule matching and command parsing. Today’s full-process autonomous Agents require complex task planning, resource scheduling, and decision-making capabilities, marking a significant leap in the intelligence level and application depth of AI Agents.
Standardization of Industry-level Agents
Standardization of industry-level Agents theoretically reflects the maturity of AI Agents in industry applications. It means that a relatively fixed and reusable framework of intelligent solutions has been formed to address the common needs and business processes of different industries. For example, in the medical field, the “Smart Medical Assistant” is backed by a deep understanding and modeling of the medical knowledge system, medical record data structure, and imaging diagnosis processes. By constructing a medical knowledge graph, linking disease symptoms, diagnostic standards, and imaging features, Agents can reason and judge based on the input of medical imaging and patient information, assisting doctors in making more accurate diagnoses. This standardized solution not only improves the efficiency and accuracy of medical diagnosis but also provides a replicable model for the intelligent development of the medical industry. In the industrial sector, the industrial Agent of the Chaos platform, based on the collection and analysis of industrial equipment operating data, uses fault prediction models (such as deep learning-based anomaly detection models) to achieve early warning and prediction of equipment failures, reducing maintenance costs. This is a typical embodiment of the standardized application of AI Agents in the industrial field, promoting intelligent and refined management of industrial production.
Upgrading Human-Machine Collaboration Models
During the WAIC live experiment, the audience commanded the Agent team to complete cross-departmental collaboration projects using voice commands, taking only one-third of the time compared to traditional methods.
In the trend of upgrading human-machine collaboration models, the underlying theoretical logic is to redefine the relationship between humans and AI. In the traditional model, humans act as operators, primarily responsible for executing specific tasks, while AI often serves as an auxiliary tool. However, as AI Agent capabilities improve, humans gradually transition to managerial roles. This requires humans to master new skills, such as optimizing prompts, which essentially involves guiding Agents to better understand task intentions through more precise language expression, involving semantic understanding and interaction theory in natural language processing. Setting task priorities requires humans to reasonably rank multiple tasks executed by Agents based on their understanding of business goals and resource constraints, drawing on priority sorting theories in project management. In the WAIC live experiment, the audience commanded the Agent team to complete cross-departmental collaboration projects using voice commands, demonstrating a new type of human-machine collaborative workflow. The Agent team, based on high-level human instructions, utilized their task planning and execution capabilities to coordinate workflows across different departments, significantly improving collaboration efficiency, validating that in the new human-machine collaboration model, reasonable division of labor and collaboration can fully leverage human decision-making capabilities and AI Agent execution capabilities, achieving a substantial increase in work efficiency.
3. Challenges and Opportunities
How Far is the Agent from Being the “Perfect Colleague”?
In the current landscape of challenges and opportunities, AI Agents still have a long way to go to become “perfect colleagues”. On a technical level, the hallucination problem is a major bottleneck—during complex tasks, Agents may make erroneous API calls due to uncertainties in understanding and reasoning with large model knowledge, which needs to be addressed through formal verification, knowledge graph semantic validation, and other mechanisms to strengthen logical verification; multi-Agent collaboration also faces challenges, as different intelligent agents have varying goals and knowledge systems, requiring the establishment of unified communication standards and collaboration protocols, combining distributed artificial intelligence, operations research, and other theories to achieve efficient collaboration. Security and ethics are also critical, as the expansion of Agent applications highlights risks such as unauthorized access to sensitive data. The establishment of the Global AI Innovation Governance Center at WAIC is promoting the formulation of ethical norms to delineate boundaries for Agent behavior, ensuring data security and social fairness. In commercial implementation, the consensus has emerged that “only closed-loop Agents have value”; logistics, customer service, and other scenarios have validated ROI, such as Xijing Technology’s logistics Agent improving efficiency and intelligent customer service reducing costs, but creative applications still need to explore business models and evaluation frameworks due to the difficulty in quantifying value and reliance on multidimensional comprehensive understanding. Behind these challenges lie significant opportunities for technological breakthroughs and industrial upgrades, driving Agents towards a more mature and reliable direction.
Agents Will Reshape Work and Life
WeChat ID丨Shu Jie Technology Sudo X