Is it harder to manage AI than to manage human employees? A former president of LinkedIn China talks about the management of digital employees.
Editor's Note
Humans and AI agents will form a new type of collaborative relationship, which requires enterprises to adopt the mindset of "digital and intelligent employee management."
The rapid development of artificial intelligence (AI) technology is influencing all aspects of society with unprecedented depth and breadth, especially bringing profound changes to enterprise operation models, business models, and even the entire social structure. As 2025 has arrived, discussions about AI agents are increasing day by day, and their potential disruptive power has attracted wide attention. In this context, it is crucial to deeply understand the core concepts, application prospects, and challenges brought by AI to grasp future development trends.
Reid Hoffman, the co-founder of LinkedIn and an early investor in OpenAI, elaborated on how artificial intelligence empowers all aspects of social life in his book Superagency. Our magazine interviewed Lu Jian, the translator of Superagency, former global vice president of LinkedIn and former president of LinkedIn China, to conduct an in - depth discussion on the impact of AI on enterprise business models, core competitiveness, and long - term trends in the business field. We hope it can provide useful references for enterprises to better understand the value and methods of AI empowerment and more effectively promote intelligent transformation.
"Superagency": The Key to Understanding AI Empowerment
When discussing the impact of AI, "Agent" is a core concept. However, simply translating it as "agent" may not fully reflect its connotation.
In the Chinese version of this book, the author - defined "Superagency" is translated as "Superagency". The reason for this choice is the discrimination of the multiple connotations of "Agency" in the history of human civilization. Different from the "agent" in daily context (such as insurance agents, travel agents), Anthony Giddens, the founder of sociology, emphasized in his structuration theory: "Agency is the ability of actors to change established social rules and the source of creativity to break through structural constraints." This philosophical speculation projected into the technical field reveals the essence of human anxiety in previous industrial revolutions - when the spinning jenny replaced the spinning wheel and when the assembly line replaced the artisan workshop, on the surface, it was the fear of unemployment, but in the deep, it was the fear of the deprivation of the subject's agency by the technical system.
Through twenty years of technological observation in Silicon Valley, Hoffman found that the particularity of the AI revolution lies in that it touches the core area of human cognition for the first time. Empirical research in the book shows that 79% of respondents are worried that AI will weaken human decision - making dominance. And the ultimate definition of "Superagency" is a response to this anxiety: empowered by AI technology, individuals and groups break through the traditional limitations of physiology, cognition, and social structure to maximize their potential.
The emergence of AI technology, especially large language models (LLMs), often triggers people's concerns about whether their own agency will be restricted. However, the real potential of AI lies in its role as an enabler of "Superagency". This empowerment is not only reflected in enhancing individual abilities but also in enabling individuals and groups to break through existing limitations and maximize their potential through technological means. This is an effect that accumulates at the social level and benefits everyone. Therefore, regarding AI as a tool to enhance human agency rather than a substitute for human subjectivity is the key perspective to understand its positive significance.
Can AI Empathize with Humans?
A common view on the nature of large models is that they are merely statistical calculations and pattern matching based on massive data, lacking real understanding and emotion. This kind of doubt is particularly prominent in application scenarios involving emotional communication, such as mental health care. For example, Rob Morris, a technology developer, used an AI assistant named KokoBot to participate in a peer - to - peer mental health support project, aiming to simulate the empathy and support of human peers. However, when its AI identity was made public, it triggered ethical disputes about "simulated empathy".
Despite the doubts, practice shows that AI's performance in specific situations may exceed expectations. Research shows that in scenarios such as doctor - patient conversations, trained large models may even score higher than human professionals in terms of empathy ability. This shows from one aspect that the criteria for judging the value of AI may not be limited to whether it has "real" emotions or consciousness, but should focus more on the effects and values it generates in practical applications. When users still feel understood and supported after knowing that their conversation partner is an AI, it itself indicates the potential of AI in assisting emotional communication.
Of course, the limitations of AI cannot be ignored, such as the "hallucinations" it may produce (i.e., generating untrue or misleading information). But this does not mean that the application of AI in fields such as mental health is worthless. The key lies in clarifying its positioning: AI is more suitable as an auxiliary tool for nursing or participating in supportive services rather than replacing professional doctors for diagnosis and treatment. At the same time, with the progress of technology and the maturity of application models, people's acceptance of AI in emotional communication is gradually increasing. Perhaps, human reaction and decision - making mechanisms are not entirely based on deliberate emotional processing, and sometimes may be based on habits or rapid reactions formed by a large amount of experience. From this perspective, the "instant" responses generated by AI through learning massive data may have some similarities with certain human reaction patterns.
AI - Native Applications and the Evolution of Business Models
The mobile Internet gave birth to native applications such as Uber and Douyin based on mobile Internet technology, creating new business models and a new group of industry leaders. What characteristics will the native enterprises in the AI era have?
From a technical perspective, the explosion of AI technology has given rise to a large number of AI - native applications. The core of these applications is that the large AI model or AI system itself is the protagonist of the product, rather than just an additional function. Typical examples include large language models that can directly communicate with users (such as ChatGPT) and AI painting tools that can generate images or artworks based on prompts (such as Midjourney, DALL·E). These applications were unimaginable before the emergence of AI technology, representing the possibility of product innovation directly driven by AI technology.
However, compared with the "native APPs" such as Uber and Douyin in the mobile Internet era, which created new business models by leveraging mobile features, the innovation of business models in the current AI era is still in its early stage. Although a large number of AI - native applications have emerged, the truly mature and large - scale replicable new business models (such as the sharing economy model of Uber) have not fully emerged. This may be because it still takes time for AI technology to move from explosion to large - scale commercial implementation, and business model innovation often lags behind technological development.
Looking forward to the future, the following trends may occur in the business models of the AI era:
1. From "the sheep pays but the pig gets shorn" to direct user payment: Traditional Internet platforms often adopt the "the sheep pays but the pig gets shorn" model, that is, charging B - side (advertisers) to subsidize the free use of C - side (users). This model relies on a large - scale user group and standardized services. However, AI - driven personalized services make the needs and experiences of each user different, which reduces the efficiency of large - scale standardized advertising placement. Therefore, in the future, more services may shift to direct user payment, provided that AI can provide high - value personalized experiences that users are willing to pay for.
2. Service as advertising, value - driven payment: Users generally have a low acceptance of advertising forms directly embedded in answers or content. However, if AI provides a continuous and high - value service (such as personal assistants, professional consultations, etc.), users are more likely to pay for it. This requires AI services to truly solve user pain points and create perceptible value.
3. Expansion of the employment and leasing models: AI agents, because they can independently complete specific tasks, are gradually regarded as "digital employees". Enterprises may pay fees according to the number of deployed agents or the amount of tasks completed, similar to the human resource dispatching or outsourcing model. On the C - side, household service robots may also be leased rather than purchased, with service providers responsible for the maintenance and update of the robots.
The growth model of AI enterprises is different from that of traditional Internet enterprises. Taking OpenAI as an example, its user growth rate far exceeds that of any technology company in history. ChatGPT reached 100 million users in only two months, while it took nine months for Douyin and four and a half years for Facebook. This shows the extremely high user acceptance speed and market penetration potential of AI technology.
However, in terms of revenue growth, current AI enterprises generally face the situation of high investment and low (or no) direct revenue, and their business models are still under exploration. Nevertheless, judging from indicators such as user scale and query volume, the growth rate of AI is still amazing.
Will AI become a new "foundation" like an operating system or a smartphone and compete with existing giants?
At present, it seems that AI is more likely to exist as a new operating system or interaction paradigm. The traditional operating system is the interface for human - computer interaction, while the future AI operating system (or agent platform) may be the back - end system for AI to independently plan and execute tasks. The way humans interact with AI will change fundamentally, shifting from command - based interaction to goal - oriented delegated interaction. This AI operating system will coexist with existing operating systems (such as Windows, iOS), but its functions and positioning are completely different.
How to Effectively Integrate Agents into Enterprise Management
AI agents represent a paradigm shift in human - machine interaction, comparable to the shift from the command line to the graphical interface or from PCs to mobile devices. It is not only a technological iteration but also a reconstruction of interaction logic and relationships. Agents have the ability to independently plan, make decisions, and execute complex tasks and are expected to become users' digital assistants or even digital avatars. Therefore, agents are not a short - term transitional stage but a long - term development trend.
Enterprises face a series of complex and urgent challenges when deeply integrating agents into core business processes, especially in decision - making links.
The primary challenge comes from the uncertainty of the technology itself. Large models are often regarded as "black boxes", and it is difficult to trace and explain their internal reasoning processes and decision - making bases. Their outputs are inherently unstable, and the same prompt may produce different results at different times or in different environments. In the execution of multi - step and long - chain tasks, the risk of the model generating "hallucinations" (i.e., fabricating facts or misinformation) will significantly increase, which poses a threat to business scenarios that require high precision and consistency.
Secondly, the blurred area of responsibility attribution poses a major management problem. When an agent independently executes a task based on its judgment and causes mistakes or even losses, the accountability mechanism is not clear - should the designer, the deploying enterprise, the operating employee, or the agent itself (which does not have subject qualification under the existing legal framework) be responsible? How to clearly define the responsibility boundary between human employees and autonomous AI systems? How to ensure that the outputs and behaviors of agents always comply with industry norms, company policies, and laws and regulations (compliance guarantee)? Currently, there are no mature solutions to these problems.
Moreover, the operational risks brought by ecological dependence cannot be ignored. The case where the API service of the programming assistant Windsurf was cut off by Anthropic after it negotiated an acquisition with OpenAI highlights the vulnerability of application providers relying on a single closed - source large - model platform for core capabilities. Although open - source models provide the possibility of avoiding vendor lock - in, for most enterprises (especially small and medium - sized enterprises), the cost of independently maintaining the development, tuning, and operation of high - performance open - source models is high, and the performance of current open - source models still lags behind that of top - notch closed - source models.
To address these challenges, leading enterprises have begun to build an "Enterprise - level AI Governance Triangle Framework": establish a strict audit mechanism to achieve full - link traceability and interpretability of key operations. Design effective alignment mechanisms to ensure that agent behaviors comply with enterprise values and business goals through prompt engineering, fine - tuning, and other means. Build a trust mechanism to alleviate the general distrust of AI outputs among employees and customers through increased transparency and technological process optimization.
In the technical architecture, adopting model - agnostic agent design and multi - cloud/multi - model strategies to avoid deep binding to a single technical platform has become a practical choice to ensure business continuity and reduce risks.
***
The empowerment potential of AI is accelerating its release in multiple strategic fields, profoundly reshaping the industrial landscape. Knowledge - intensive fields are the first to be affected. The efficiency of white - collar work scenarios such as software development (automatic code generation, function completion, bug detection, test case generation) and marketing (personalized copywriting creation, user portrait analysis, advertising content production) is experiencing exponential growth, and at the same time, it lowers the entry threshold for professional skills.
The education field is undergoing a revolutionary change. AI - driven personalized learning platforms can dynamically adjust teaching content and paths according to each student's understanding ability, learning rhythm, and interest preferences, providing a technological foundation for the large - scale implementation of the educational ideal of "teaching students in accordance with their aptitude".
The professional service field is also being upgraded. Applications such as assisted generation of legal documents and high - precision real - time multi - language translation (as demonstrated by the translator's experience of significantly improving efficiency with the help of AI when translating Superagency) have significantly improved the speed and quality of services. At the enterprise operation level, the response speed and accuracy of customer service (intelligent customer service) and sales support (lead analysis, sales pitch suggestions) have achieved a leap - forward improvement.
However, along with the efficiency improvement, there is a profound impact on the employment structure. The manpower demand for knowledge - intensive positions may face a structural reduction. In contrast, in the field of precision manual labor that requires high - level adaptability to complex environments, fine hand - eye coordination, and immediate physical feedback (such as top - notch surgical operations), the pressure of being replaced by AI automation in the short term is relatively small, and it is more likely to move towards a deep human - machine collaboration model (such as robot - assisted surgery). However, it is worth noting that with the continuous breakthroughs in the perception accuracy, motion control, and environmental adaptability of robot technology, these fields once regarded as exclusive human fortresses may also be gradually penetrated by more intelligent and reliable machine systems in the future. For example, in surgical operations, robot arms can theoretically achieve stability and precision far beyond the physiological limits of humans.
Humans and AI agents will form a new type of collaborative relationship - humans are responsible for setting goals, providing value judgments, and supervision, while AI agents are responsible for efficient execution, information integration, and providing suggestions. This collaboration requires enterprises to adopt a new thinking framework when deploying and managing agents. This means that enterprises should not only tolerate the current uncertainties of the technology (such as hallucinations and output fluctuations) but also proactively establish a comprehensive AI governance framework, covering ethical guidelines, responsibility division, safety audits, and continuous evaluation mechanisms. Only in this way can we truly enter a new era of human - machine collaborative evolution and jointly release potential.
Interviewee: Lu Jian, Former Global Vice President of LinkedIn, Former President of LinkedIn China, and Global Business Chief Consultant of CGL Group
This article is from the WeChat official account "China Europe Business Review". Author: Qi Qing. Republished by 36Kr with permission.