Intelligent agents are starting to receive “employee IDs” too. On August 31, at the Zhangjiang Science Gate Model Power Community, the intelligent agent identity code node management platform officially launched, with 30 organizations participating in node issuance. More and more intelligent agents will possess recognizable, manageable digital identities.
But as Agents evolve from tools on a personal computer into “digital colleagues” capable of cross-system invocation and participation in business processes, a question more practical than technology emerges: who assigns them tasks, who accepts their results, and who bears responsibility when problems arise? This question tests not only the capabilities of intelligent agents, but also the management capabilities of organizations.
Figure 1: Zhangjiang Science Gate Model Power Community, the identity code node management platform officially launches
At the event site, the Shanghai Artificial Intelligence Industry Association and the China Electronics Standardization Institute, among other organizations, worked in concert to comprehensively advance the promotion and publicity of AIP agent interconnection applications, injecting new momentum into the industry's digital transformation. Presenters likened the intelligent agent identity code to a “digital ID card,” breaking down the entire process step by step, from identity application and material verification to cross-entity interconnection and invocation.
With an identity code, an intelligent agent effectively gains an “employee ID” within the organization. But an employee ID is only the starting point. In the past, “one person, one tool” depended on individual choice; now, behind one person there may stand a team of intelligent agents—who assigns tasks, who accepts results, who bears responsibility? Organizations must redraw the map of division of labor.
I. First, identity: intelligent agents officially enter the organization's “roster”
Why does identity come first? Intelligent agents are moving from “answering questions” to “getting things done,” and requiring AI to “get things done, make things easy to do, and do them well” requires first clarifying: identity (who am I), agency relationship (for whom do I act), and permission boundaries (what do I do), so it can become the AI era's “intelligent butler.” According to the series of national standards on agent interconnection, one identity code precisely corresponds to one intelligent agent.
Rules landed intensively this summer. On July 17, at the 2026 World Artificial Intelligence Conference held in Shanghai, the Cyberspace Administration of China, together with relevant parties, jointly proposed the Global Cooperation Initiative on Agent Mutual Trust, Interconnection and Interoperability, calling for increased R&D investment in “foundational common technologies such as agent identity identification and mutual recognition of collaboration”; the concurrently released International Action Plan for AI Ethical Governance proposed building differentiated risk handling rules around factors such as “tool invocation permissions and autonomous decision-making capability,” and clarifying the “boundary of rights and responsibilities” among developers, deployers, and users.
Domestic policy provides synchronized guidance. The Implementation Opinions on the Standardized Application and Innovative Development of Intelligent Agents, issued by three departments including the Cyberspace Administration of China, proposed exploring the establishment of agent registration platforms offering services such as “digital identity management, search and discovery, and capability declaration.” The platform launched in Zhangjiang is practicing this path—ensuring that every intelligent agent “checks in” with a name, a surname, and a traceable record.
Shanghai is also accelerating its “filing” layout. The “Digital Shanghai” Construction 15th Five-Year Plan released in August made clear the goal to “build a government affairs agent cockpit, achieve workspace and operational observation, and promote efficient multi-level collaboration among people, agents, and systems”—intelligent agents must not only “have names,” but also be “visible and controllable.”
Identity codes solve “being recognizable,” organizational authorization solves “being controllable,” and role-based division of labor is what solves “being well used.” Between digital identity and digital employee lies a whole set of organizational management systems.
II. Then divide the labor: human-machine partners, one role one assistant
Once an Agent's identity is established, the next step is defining responsibilities. From assisting a single role, to taking on complete processes, to letting business personnel directly dispatch professional capabilities, intelligent agents are changing how enterprises organize and invoke internal capabilities. SAIC's Fanyi Shangxing has cumulatively launched more than 100 digital employees; after the production planning assistant outputs a simulated plan, it is “submitted to the planner for review.” Dr. Chen Ning, R&D lead of the digital employee project, emphasized: “The essence of a digital employee is a virtual collaborator capable of taking on the full responsibilities of a position, focusing on a single role to achieve an end-to-end closed loop.” The line of division further extends to business judgment. For example, SenseTime's Xiaohuxiong lets business personnel “only need to input one sentence of natural language” to complete data association, trend analysis, and report generation. People are freed from fetching data and can focus their energy on decision-making. Whether the division of labor is clear depends not on the number of assistants, but on whether ownership of every task segment is explicit: which steps are handed to agents, and which must be checked by humans.
Figure 2: Fanyi Shangxing's AI production planning assistant
III. The human-machine mixed formation: how to implement task assignment, acceptance, decision-making, and accountability?
Organizational forms are evolving in step. Employees can carry multiple intelligent agents working in coordination; personnel dynamically combine around tasks, and disband and recombine once tasks are completed. Managers' new responsibility lies in: designing human-machine division of labor and covering for the results.
In traditional hierarchical organizations, tasks are passed down layer by layer along departments and ranks; matrix organizations allow personnel to collaborate across departments, but remain constrained by relatively fixed reporting relationships. Organizational intelligence goes further by shifting organizational units to tasks and roles: one employee can dispatch multiple intelligent agents, professionals from different fields temporarily form teams around projects, and resources are reconfigured after tasks end. Organizational intelligence no longer has only fixed superior-subordinate relationships; it also possesses collaboration networks that change with tasks.
Task assignment must be led by those who understand the business. Guotai Haitong aims to rely on AI to drive organizational structure reform, business process innovation, and business model upgrading; its AI applications already cover nine major business areas and 260 specialized scenarios, with more than 1000 business agents deployed; over 500 employees participated in building them, including more than 100 frontline business personnel.
Acceptance requires rules to follow. People's Daily Online's No. 001 AI digital employee is positioned as an “office digital collaboration specialist,” with an assessment system that clearly specifies “strictly adhering to authorized red lines, with task completion rate, acceptance pass rate, and operational compliance as core indicators.”
Final decisions still rest with humans. The Implementation Opinions on the Standardized Application and Innovative Development of Intelligent Agents makes clear the need to clarify the reasonable boundaries and permissions of “decisions reserved for the user alone, decisions requiring authorization, and decisions made autonomously by agents,” ensuring users enjoy “the right to know and the final decision-making power” over agents' autonomous decisions. Tasks can be delegated, but permissions must be tiered; decisions can be authorized, but responsibility cannot be left hanging.
Accountability hinges on building good mechanisms, choosing the right people, and fostering good culture. Adding an “approve” button is simple; the difficulty lies in making responsibility substantive, institutionalized, and personified. When a task is completed through a relay of multiple agents, risk may also propagate along the invocation chain. One agent initiates a task, another invokes the system, and a third performs the operation—once unauthorized access or an error occurs, which link should be traced? This requires enterprises not only to establish identity authentication and permission controls, but also to retain task logs, operation records, and manual approval nodes, so that every delegation has a source, every operation has boundaries, and every result has a responsible party. Leading a human-machine mixed AI formation brings not only extra efficiency, but also management responsibility.
From issuing identity codes to agents, to defining full responsibilities for digital employees, to rebuilding the management mechanisms of human-machine collaboration, every step of AI's entry into organizations is generating new demand for professional services. Consulting and planning, process reengineering, system integration, operations management, and security governance will together constitute the service chain for the large-scale application of intelligent agents.
This also echoes the direction of the MIIT's special action to cultivate artificial intelligence application service providers. From October 23 to 25, the 2026 Global Developer Pioneer Conference will be held at the Zhangjiang Science Hall in Shanghai. As more developers, application enterprises, and professional service providers gather here, Shanghai will also host a concentrated exchange on how agents can truly integrate into organizations and become productive forces.
Issuing an “employee ID” to an Agent is only the first step of organizational change. The next step is to answer anew an old management question: how do people divide labor, how do things get done through collaboration, and how is responsibility implemented.