When you say to your phone, "Plan a graduation trip to Xiamen for me next week", it can not only work out a detailed travel guide, but also automatically compare prices according to your budget, book flights and hotels, and push electronic itineraries to you directly. This leap from a vocal advisor to a hands-on executor is exactly the subversive experience brought by AI Agents.
Recently, the Cyberspace Administration of China, the National Development and Reform Commission and the Ministry of Industry and Information Technology have jointly issued the Implementation Opinions on Standardized Application and Innovative Development of AI Agents, drawing a clear security bottom line and innovation space for the high-quality development of China’s AI agent industry. Relevant experts said that the document puts forward 19 typical application scenarios covering five major dimensions, presenting a clear roadmap specifying the application scope, usage modes and expected outcomes of AI agents.
19 Application Scenarios Covering Core National Industries and People’s Daily Lives
An AI agent is an intelligent system equipped with autonomous perception, memory, decision-making, interaction and execution capabilities. In the past, people used AI just like interviewing consultants — users ask questions and AI gives replies, while users still have to finish all execution work on their own. Today, AI agents are evolving into digital workers in all walks of life, lifting work efficiency to an unprecedented level.
The AI agent industry is now in a stage of explosive growth. According to the latest forecast by Gartner, 40% of enterprise applications will embed task-based AI agents by 2026, and the market size of enterprise-level AI agents in China will exceed 48 billion yuan. China has become one of the most vibrant markets for AI agent commercial implementation worldwide.
The Implementation Opinions divide AI agent application scenarios into five major categories including scientific research, industrial development, consumption stimulation, people’s wellbeing and social governance, involving a total of 19 typical scenarios, covering all fields from major national core equipment to ordinary daily life.
In scientific research, priority is given to developing agents for research theoretical deduction and simulation. Software development agents will be cultivated to improve full-process development capabilities including demand analysis, architecture design, code generation and testing.
Industrial development covers a wide range of fields including intelligent manufacturing, energy resources, transportation, agricultural production and financial services. It specially promotes the integration of AI agents with CNC machine tools, industrial robots and automatic production lines, and supports the research and application of agents for power dispatching, traffic safety supervision, agricultural services and financial risk control.
In terms of consumption boosting, it promotes the coordinated development of AI agents together with mobile phones, computers, vehicles, smart home facilities, wearable devices and consumer-grade robots, and accelerates the development of content creation agents for literature, music, painting, audio-visual works and performing arts.
For people’s wellbeing, applications are encouraged in education, medical care, human resources and information services. It explores intelligent tools for courseware generation, homework correction and academic situation analysis, and optimizes medical auxiliary agents for medical image analysis, disease diagnosis reasoning and customized treatment plan formulation.
In the field of social governance, applications are supported in government services, judicial services, public security, urban governance and bidding. It fosters agents for monitoring and early warning, emergency disposal, rescue scheduling and collaborative governance, and realizes full-process intelligent management of bidding activities.
Faced with this extensive and well-defined scenario list, interviewed experts believe that 2026 will be a critical turning point for AI agents to shift from technological exploration to large-scale practical deployment.
"The AI agent industry will enter a phased development cycle in the next three years," said Zou Debao, Executive Deputy General Manager of the AI & Big Data Research Center at CCID Consulting. "2026 will witness an outbreak of pilot projects, with professional AI agents in vertical scenarios such as customer service, finance and office work achieving large-scale implementation, driving market growth rate exceeding 60%. 2027 will enter the stage of expansion and integration, with industrial standards and regulatory systems gradually improved and compliance boundaries clarified, and the industry will evolve from single independent agents to A2A multi-agent collaborative networks. By 2028, the enterprise adoption rate of AI agents will exceed 70%, making them standard digital infrastructure."
He Jianghao, researcher at China Machinery Industry Information Research Institute, said that for manufacturing enterprises, apart from facilitating internal intelligent upgrading, manufacturing equipment such as industrial machine tools, industrial robots, testing devices, agricultural machinery and transportation equipment can all be equipped with AI agents, upgrading traditional standalone equipment into integrated systems of intelligent hardware plus intelligent services. Meanwhile, it will open up brand-new market space in downstream industries including energy, transportation, agriculture, medical care, public security and urban governance.
Dual-drive Development of Progress and Security
For the first time at the national level, the Implementation Opinions clarifies the boundary of decision-making authority between AI agents and users. It defines reasonable scope and required permissions for three decision-making modes: exclusive user decision-making, user-authorized decision-making and autonomous decision-making by AI agents.
Pan Helin, member of the Information and Communication Economy Expert Committee under MIIT, pointed out that users must retain the ultimate right to know and the right to halt operations under any mode. This rule settles the long-standing confusion over authority division between users and AI agents.
In terms of governance framework, the document advocates prudent tiered management of AI agents. For sensitive fields and key industries, competent cyberspace authorities together with industrial regulators will define accessible scenarios, and implement filing, inspection and defective product recall mechanisms in accordance with laws, regulations and security standards. For low-risk fields such as daily entertainment and office affairs, optimized evaluation tools will be popularized to realize efficient governance via self-compliance inspection, information reporting, platform management and industrial self-regulation.
Notably, the document proposes to explore the establishment of a unified AI agent registration platform. According to Zou Debao, this is equivalent to issuing a digital ID card for each compliant AI agent, enabling users to inquire about its developer, data access authority, security level and compliance status.
Pan Helin stated that the policy has released three core orientations: first, adhering to dual-drive and balanced development of progress and security; second, taking practical applications to lead technological innovation and integrate technology research with real-scene implementation; third, pushing forward standardized development to deepen in-depth industrial integration and coordinated growth.
Breaking the Dilemma of Wide Praise yet Low Adoption
Despite promising prospects, the AI agent industry is still trapped in the awkward situation of high recognition but sluggish landing. According to McKinsey data, although 62% of enterprises are conducting AI agent trials, less than 10% have realized large-scale deployment. Technical fragmentation, uncertain security performance and poor compatibility with existing workflow have become three major obstacles restricting industrial popularization.
"AI agents will achieve geometric growth in the next three years with rapidly expanding user base and booming industry scale, similar to the explosive development trend of new energy vehicles in the past. A large number of traditional software and hardware products will take the initiative to integrate with AI," Pan Helin commented. Enterprises that resist AI agent transformation will face enormous competitive pressure.
Targeting these industry pain points, the Implementation Opinions puts forward measures to smooth supply-demand links, promote high-level interaction between R&D side and demand side, and build a market-driven and internally-motivated industrial ecosystem for AI agents.
Specific measures include fostering open-source innovation vitality, promoting compatibility adaptation between AI agents and open-source chips, operating systems and large models, building industrial cooperation platforms to jointly carry out generic technology research, standard formulation and evaluation certification, establishing application promotion channels such as AI agent software stores and industrial information platforms, opening up more pilot scenarios in industrial clusters and key sectors, and actively cultivating global industrial ecology while guiding enterprises to improve overseas compliance layout.
Relevant officials from the Cyberspace Administration of China said that in the next stage, relevant departments will focus on key links including AI agent technological R&D, scenario opening and security governance, improve supporting policies, form joint working forces to ensure the implementation of key tasks, and strengthen dynamic monitoring, phased implementation and flexible policy adjustment for the standardized application and innovative development of AI agents.
Reporter: Du Zhuang
Source: China Development and Reform News
当你对着手机说“帮我策划一场下周去厦门的毕业旅行”，它不仅能制定出一份详尽的攻略，还能直接根据你的预算自动比价、预订机票酒店，并把电子行程单推送到你面前。这种从“动嘴的参谋”到“动手的打工人”的跨越，正是智能体（Agent）带来的颠覆性体验。
近日，国家网信办、国家发展改革委、工业和信息化部联合印发《智能体规范应用与创新发展实施意见》（以下简称《实施意见》），为我国智能体产业的高质量发展划定了“安全底线”与“创新空间”。相关专家表示，《实施意见》从五大方向提出19个典型应用场景，为智能体“往哪用、怎么用、用出什么效果”给出了一份清晰的路线图。
19个应用场景贯通“大国重器”与民生日常
智能体是具备自主感知、记忆、决策、交互与执行能力的智能系统。过去，我们用AI像是在“面试”顾问，你问他答，执行还得靠自己；如今，智能体正化身为各行各业的“数字打工人”，将效率提升到前所未有的高度。
当前，AI智能体行业正进入爆发式增长阶段。Gartner最新预测显示，2026年将有40%的企业应用嵌入任务型AI智能体,中国企业级AI智能体市场规模突破480亿元。中国已成为全球最活跃的智能体落地市场之一。
《实施意见》将智能体应用场景划分为科学研究、产业发展、提振消费、民生福祉、社会治理五大方向19个典型场景，覆盖从“大国重器”到寻常日用的全链条。
科学研究领域重点布局研发理论推演、模拟仿真等智能体。发展软件开发智能体，提升需求分析、架构设计、代码生成与测试等全流程开发能力。
产业发展覆盖较广，包括智能制造、能源资源、交通运输、农业生产、金融服务五大行业，特别提到促进智能体与数控机床、工业机器人、自动化产线等融合，发展电力调度智能体，研发交通安全监管、农业服务、金融风控等智能体。
提振消费方面，推动智能体与手机、电脑、汽车、家居、可穿戴、消费级机器人等终端设备协同发展，同时研发文学、音乐、绘画、视听、演艺等内容创作智能体。
民生福祉方面，鼓励覆盖教育教学、医疗健康、人力资源、信息服务四大方向，探索课件生成、作业批改、学情分析等智能体，提升医学影像分析、疾病诊断推理、定制化诊疗方案生成等医疗辅助智能体性能。
社会治理领域，支持覆盖政务服务、司法服务、公共安全、城市治理、招标投标五大方向。探索监测预警、应急处置、救援调度、协同治理等智能体。实现招标投标活动全链路智慧管理。
面对这份覆盖广泛、指向明确的场景清单，受访专家认为，2026年将是智能体从技术探索迈入规模化落地的关键转折点。
“未来三年智能体产业将迈入梯度发展周期。”赛迪顾问人工智能与大数据研究中心常务副总经理邹德宝对记者表示，2026年为试点爆发期，客服、金融、办公等垂直场景专业智能体迎来规模化落地，市场增速将超60%；2027年将进入扩张整合期，行业标准与监管体系逐步完善、合规边界愈发清晰，产业将从单点智能体演进至多智能体A2A协同网络；2028年将步入成熟渗透期，智能体企业普及率突破70%并成为数字化标配。
机械工业信息研究院研究员贺疆澔告诉记者，对于制造业企业来说，除助力企业实现自身智能化升级外，工业母机、工业机器人、检测设备、农机及交通装备等制造产品，均可搭载智能体，从传统单机设备升级为“智能装备+智能服务”系统。同时还将打开能源、交通、农业、医疗、公共安全、城市治理等下游行业全新市场空间。
发展和安全双轮驱动
《实施意见》首次从国家层面明确了智能体与用户之间的决策权限边界。《实施意见》指出，厘清仅限用户本人决策、需由用户授权决策和智能体自主决策等各种决策方式的合理边界及所需权限。工信部信息通信经济专家委员会委员盘和林告诉记者，无论哪种模式，用户都必须保留最终的知情权和“叫停权”，这一规定解决了智能体和用户决策权限的分界问题。
在治理框架上，《实施意见》指出要审慎稳妥开展智能体分级治理。对于敏感领域及重点行业，由网信部门联合行业主管部门确定开放场景，根据相关法律法规、监管要求和安全防护标准，实行备案、检测、问题产品召回等管理措施。对于部分生活娱乐、日常办公等低风险领域，完善智能体评估测试工具，通过合规自测、信息报告、分发平台管理、行业自律等实现高效治理。
值得注意的是，《实施意见》要求，探索建立智能体注册平台。邹德宝说，这就相当于为每个合规智能体发放类似“数字身份证”的标识，方便用户查询其开发者、数据权限、安全等级及合规状态。
盘和林认为，政策释放了三大核心导向：一是释放了发展和安全双轮驱动，均衡发展的导向；二是释放了应用牵引技术发展，技术和应用落地结合的导向；三是规范化发展，促进智能体和产业深度融合，协同发展的导向。
破解“叫好不叫座”难题
尽管前景广阔，但当前的智能体行业仍面临“叫好不叫座”的尴尬。据麦肯锡数据显示，虽有62%的企业在试验智能体，但真正实现规模化部署的不到10%。技术碎片化、安全性存疑、难以深度融入工作流，成为横亘在企业面前的三座大山。
“未来三年，智能体将实现几何级增长，增长速度将越来越快，用户数将爆发，产业将进入成长期，类似于当年锂电车爆发的增长状态，很多传统软件和硬件将主动和AI融合发展。”在盘和林看来，拒绝智能体或者掉队的企业，将面临巨大竞争压力。
针对这些痛点，《实施意见》指出，畅通供需渠道，促进研发侧、需求侧高水平互动，形成市场牵引、内驱发展的智能体产业生态。比如培育开源创新力量，开展智能体与开源芯片、开源操作系统、开源大模型兼容适配。搭建产业协作平台，协同产业链上下游开展智能体共性技术研发、标准制定、评估认证等工作。构建应用推广渠道，推动建立智能体软件商店、行业供需信息发布平台。推进重点场景开放，在产业集聚区、重点行业、重点领域开展智能体应用试点。积极培育全球生态，引导相关企业做好海外合规建设。
国家网信办相关负责人表示，下一步，国家网信办、国家发展改革委、工业和信息化部将聚焦智能体技术研发、场景开放、安全治理等关键环节，完善配套政策，形成工作合力，推动重点任务落实落地。同时，加强智能体规范应用与创新发展的监测评估、滚动实施和动态调整。（中国发展改革报社记者 杜壮）
返回首页