With the continuous breakthrough of large-scale models, the popularity of AI intelligent agents continues to increase. Since the beginning of this year, from startups to tech giants, they have accelerated their layout and continuously expanded the application scenarios of intelligent agents. Beijing, Shanghai and other places have successively introduced policies to inject new momentum into the development of intelligent agents. What is an AI agent and what can it do? Why are AI companies deploying intelligent agents one after another? What issues should we pay attention to when developing intelligent agents? Multi agent layout intelligent agents are generally considered as advanced artificial intelligence systems that can autonomously perceive, think, and act in specific environments. They can understand, learn, and reason to perform complex tasks and make decisions. At present, intelligent agents have been applied in scenarios such as content creation assistants, knowledge Q&A assistants, intelligent assistants, and AI search. In a broad sense, AI agents are intelligent applications that can think deeply, plan autonomously, make decisions, and execute deeply. However, in the implementation process, each manufacturer and product will make different changes and combinations according to the user population and usage scenarios. ”The person in charge of Quark, an AI application under Alibaba, said. Since the beginning of this year, various companies have successively launched intelligent agent products that gather their respective advantages and strengths. Monica, a domestic big model team, has released the universal AI intelligent agent product Manus and announced a formal strategic cooperation with Alibaba's Tongyi Qianwen team. Zhipu AI releases AI intelligent agent product AutoGLM Contemplation, promoting AI intelligent agents to enter a new stage of "thinking while doing". Quark creates a super intelligent agent in the form of an "AI super box", which has advantages in fields such as search, browser, scanning, and question taking. Lenovo Group has successively launched "urban super intelligence" in Mount Wuyi, Yichang and other cities to provide customized "AI+" engines for thousands of industries and users. ByteDance Agent product "Button Space" recently opened internal test. What can current intelligent agents do? Li Wei, Deputy Director of the Cloud Computing and Big Data Research Institute at the China Academy of Information and Communications Technology, stated that currently, in terms of enterprise level applications, intelligent agents can be applied to further practice cost reduction, efficiency improvement, and data-driven decision-making for enterprises. For example, in the logistics industry, intelligent agents combined with warehouse robots can achieve automatic sorting and path planning, reduce sorting error rates, and improve warehouse throughput and circulation efficiency; In the field of human resource management, intelligent interviewers deployed in the cloud can flexibly scale and process massive video interview data, using cloud GPU accelerated multimodal analysis to evaluate candidate matching. This not only improves HR screening efficiency by 80%, but also continuously optimizes evaluation models through cloud based continuous learning mechanisms to avoid human bias. In terms of consumer applications, Li Wei believes that intelligent agents focus on providing scenario based services and personalized experiences for individuals. For example, the intelligent marketing engine in the fast-moving consumer goods industry runs in the cloud, dynamically generating personalized recommendation strategies and promotional content based on real-time analysis of user behavior data streams; The cloud edge collaborative agent in the smart home scene uploads the user habit data to the cloud training personalized model, and then through the edge computing node, real-time control the home environment equipment, automatically optimize the home environment in terms of temperature, lighting, ventilation, etc. In short, the development of intelligent agents will reshape the relationship between humans, AI tools, and tasks. Alibaba Group Vice President Wu Jia recently stated that many people say that in the era of AI, they want to redo all products. In fact, the true meaning is to redo these tools in order for AI to better use them. "AI uses tools, while people use AI. Driven by the two major factors of technology iteration and market demand upgrading, entrepreneurial companies and Internet giants have successively laid out agent products and agent development platforms. The bidirectional resonance between breakthroughs in underlying technologies and the release of industrial value is catalyzing the accelerated evolution of intelligent agents from technological concepts to commercial implementation. According to CCID Consulting's calculations, the global AI intelligent agent market is expected to maintain a compound annual growth rate of over 40% over the next five years. The policy of Dongfeng accelerates the industrial process. On April 8, the Beijing Municipal Bureau of Economy and Information Technology issued the Notice of Beijing on Supporting Information Software Enterprises to Strengthen the AI Application Service Capability (2025), proposing to support the development of general agents, support innovative agents to develop cross domain, multi task, self planned general agents, give priority to coordinating computing support for general agents that have obtained the online batch number of generative AI products and services, and are on the shelves of various application stores for the first time, and give up to 30 million yuan of support for computing and model costs in operational services. On April 21st, the Shanghai Municipal Commission of Economy and Information Technology issued a notice on the unveiling and leadership of the new generation of general artificial intelligence innovation tasks in 2025, which mentioned exploring the organic collaboration and heterogeneity of numerous intelligent agents in complex open environments, and achieving sustainable emergence of swarm intelligence, tackling key technologies such as multi-agent systems and optimization decision-making technology, unmanned cluster system technology, swarm intelligence software technology, and swarm intelligence federated learning technology. In terms of personal applications, Quark utilizes its technological capabilities to help meet the diverse needs of users in various scenarios. Quark uses a minimalist 'AI super box' to become a versatile personal assistant. ”Wu Jia stated that in the future, humans will no longer need to directly use tools such as search, but will directly hand over complete task instructions to AI. AI will think, execute, and complete the final task delivery. Smartphones, with the support of intelligent agents, also understand humans better. It is reported that the Honor YOYO intelligent agent is currently capable of understanding 600 requirements and intentions, memorizing 950 personal habits, and planning 270 complex tasks. Operations such as ordering coffee, querying or canceling automatic renewals, and generating ID photos can all be completed in one sentence. In terms of empowering enterprises, Volcano Engine provides multiple AI intelligent agent solutions to customers in multiple industries through the HiAgent platform. In the financial industry, intelligent agents are experimenting and exploring in scenarios such as customer service, wealth management, and exhibition assistants; in the medical industry, intelligent agents are experimenting and exploring in scenarios such as intelligent guidance, in-hospital services, and patient health management; in the education industry, intelligent agents are experimenting and exploring in scenarios such as campus communication, teacher courseware assistants, and research assistants; in the manufacturing industry, intelligent agents are experimenting and exploring in scenarios such as equipment maintenance assistants, research and development assistants, and embodied intelligence. "said the person in charge of Volcano Engine. In the field of smart cities, Lenovo Group recently launched the "urban super agent" in Mount Wuyi, Yichang and other cities, and adopted the "1 × N agent scheme", that is, the mode of one super agent working with agents in multiple fields, to deliver AI capabilities to all walks of life and citizens and tourists in the city, to achieve comprehensive intelligence from government affairs to people's livelihood and industries, and to mix small scenes to form a big scene to promote the smart development of the city. By leveraging their strengths and weaknesses, intelligent agents not only retain the powerful data processing capabilities of large models, but also introduce self boundary judgment, active perception, complex task decomposition, and memory mechanisms, making artificial intelligence more closely related to the practical application scenarios of human intelligence, "said Dai Wei, Senior Vice President of Lenovo Group and General Manager of China Solution Service Business Group. Han Zizhe, an analyst at the Big Data and Artificial Intelligence Industry Research Center of CCID Consulting, said that the autonomous decision-making ability and tool calling ability of intelligent agents make AI no longer an isolated technical module, but a "productivity unit" that can be deeply embedded in enterprise operation systems. It is expected to solve the problems of fragmentation and low input-output ratio in traditional AI application scenarios, which will promote the application of AI technology on the industrial side. When the industry rushes towards intelligent agents, technological challenges and various risks also follow. The first thing to be wary of is' pseudo intelligent agents'. Li Wei stated that with the increasing popularity of the concept of intelligent agents, a group of "pseudo intelligent agents" have emerged in the market under the banner of intelligent agents. Many companies package or label traditional technologies and existing products as intelligent agents, and promote them through marketing strategies to mislead users. Secondly, it is necessary to pay attention to technical bottlenecks. Han Zizhe stated that attention should be paid to the technological bottlenecks that affect the cognitive reliability of intelligent agents, including machine illusions, decision black boxes, and other issues, which are the technical foundation for ensuring trustworthy decision-making of intelligent agents. Li Wei stated that AI agents highly rely on large language models, but their accuracy in generating content and contextual understanding still cannot meet the application requirements of production level scenarios. The existence of machine illusion problems may lead to decision-making errors in high-risk fields such as finance and healthcare. Once again, the issue of cost also requires attention. When using intelligent agents to perform large and complex tasks, the consumption of tokens significantly increases. For example, Manus controls the inference cost to about one tenth of the inference cost of OpenAI's DeepReach agent, and each task still requires 500000 to 2 million tokens to be consumed. "As the number of users and user usage increases, the exponential growth of computing resources is an inevitable cost input for the implementation of agents. However, the computing power demand of very large scale GPU clusters is far beyond the scheduling capacity of traditional data centers, and it is necessary to rely on AI Cloud, Computing Internet and other new computing interconnected scheduling architectures to achieve efficient and low-cost computing calls. ”Li Wei said. Finally, attention should be paid to the issues of technology ecology and collaborative standardization. The standardization of technology ecology and collaboration is the key to determining the large-scale implementation and cross scenario collaboration capabilities of AI intelligent agents. However, the current protocol standards have not yet formed a unified standard, and the collaborative efficiency has not been effectively improved. Han Zizhe believes that industry users need to pay attention to the construction of high-quality internal datasets, accumulate data based on business scenarios, drive the optimization of intelligent agents with high-quality data, and then feed back the improvement of business value, thus forming a virtuous closed loop. Li Wei suggested that the industry urgently needs to build a unified standard similar to the Internet protocol cluster, optimize the multi-agent architecture, release the synergy potential, and jointly promote AI agents from "single point breakthrough" to "ecological prosperity". The development of AI intelligent agents is highly in line with the digital transformation needs of China's economy and society, which is a critical period from technical verification to large-scale implementation. Looking towards the future, it is necessary to coordinate technological innovation and governance system construction. While consolidating the foundation of independent and controllable technologies, we should promote collaborative research and development among industry, academia, and research institutions, explore sustainable models for the deep integration of intelligent agent technology and the real economy, and inject lasting momentum into the leap of productivity in the digital age. (New Society)