Li Zhenhua, Zhang Yun/Wen

Over the past 20 years, the most important contribution of the Internet is to connect information and achieve efficient matching between supply and demand. It makes it easier for consumers to find goods and services, enables businesses to reach customers more efficiently, and has also spawned a group of globally influential enterprises.

Today, intelligent agents are pushing digitization to a new stage, not just answering questions and matching information, but understanding human intentions, breaking down tasks, calling tools, coordinating resources, organizing actions, and autonomously completing tasks.

Throughout the technological revolutions, major technological breakthroughs will give rise to new business models and have a profound impact on market structure and patterns.

Entering the era of intelligent agents, reducing transaction costs remains the core of market development. What we need to consider is: what changes will occur in the structure and pattern of the market when the process of value creation and exchange shifts from being people-centered to involving both humans and intelligent agents? What are the core market frictions and transaction costs that need to be addressed? How can enterprises find new roles and value moats? How will technological potential be transformed into economic growth and broader social welfare.

Intelligent weight structure decision-making power

In the past, the Internet has organized a lot of information, but people are still the subject of decision-making and action. The intelligent body constitutes the decision-making power of economic activities, moving from a people-centered approach to the joint participation of humans and intelligent agents: humans are only responsible for setting goals, setting boundaries, and making key confirmations, while intelligent agents are responsible for obtaining information, comparing decisions, organizing resources, and coordinating communication. As a result, the functions, structure, and boundaries of the market will undergo changes.

Firstly, the unit of commercial activity has shifted from trading goods and services to task trading, and the function of the market has shifted from organizing information to organizing action.

In the past, when arranging a business trip, users needed to purchase air tickets, book hotels, and arrange transportation separately; In the future, it may only be necessary to propose "arranging an efficient, comfortable, and within budget business trip", and intelligent agents can coordinate multiple services to complete tasks.

In business operations, intelligent agents can also leverage data, tools, and professional services to organize a complete action process around the goals of reducing supply chain costs and conducting market research. What users purchase is no longer just goods and services, but task results formed around specific goals; Merchants' operations will also evolve from providing individual SKUs to delivering complete solutions.

Secondly, the market structure has shifted from a stable bilateral market to a dynamic multilateral action network.

The Internet platform mainly connects consumers and businesses to form a relatively stable bilateral market. In the intelligent agent economy, a task may simultaneously connect user agents, merchant agents, model vendors, developers, data and tool service providers, payment institutions, and other professional agents.

The objects of market transactions not only include physical goods and services, but also digital resources such as models, data, tools, and professional capabilities. Different entities dynamically connect and collaborate around specific tasks, and continuously break down and reorganize as tasks change. The market has transformed from a relatively stable supply and demand connection to a multi-party collaborative action network.

Thirdly, the digital boundaries of the market have been further expanded, shifting from a focus on consumer interconnectivity to a broader range of industrial interconnectivity.

An important reason for the rapid development of consumer Internet is that goods and services are relatively standardized and easy to display, compare, pay and evaluate. However, industrial activities such as enterprise procurement, supply chain collaboration, legal compliance, software development, and professional consulting require an understanding of complex backgrounds, handling multiple constraints, and coordinating multiple entities, resulting in long-term high digital barriers.

Intelligent agents are able to understand natural language objectives, call upon different systems and professional capabilities, gradually breaking down complex services that were previously difficult to standardize into discoverable, callable, composable, and verifiable action processes, promoting digitalization to further enter the deep scene of enterprise production and operation and industrial chain collaboration.

Is it credible for core market friction to turn into action

The systematic reduction of search, negotiation, and contracting costs by intelligent agents has improved the efficiency of supply and demand matching, but it has also brought new problems - how to trust the actions of intelligent agents.

As more and more consumption, production, and business activities are entrusted to intelligent agents, the principal-agent relationship becomes the fundamental structure of microeconomic operation. How to make the actions of intelligent agents trustworthy has become the most core and critical market friction that needs to be addressed in the intelligent agent economy.

The potential principal-agent friction brought by intelligent agents includes two levels.

Firstly, the trust between humans and intelligent agents, that is, whether humans can trust the intelligent agent to accurately understand the target and act faithfully and diligently within the authorized scope.

People expressing their needs through natural language often only provide general goals and it is difficult to exhaust all conditions. For example, "choosing the most suitable supplier" may involve price, quality, delivery stability, compliance risks, and long-term cooperative relationships. Intelligent agents need to make autonomous judgments in areas that are not explicitly stated, which may result in intention understanding bias and autonomous decision-making bias, and even exceed the client's true expectations in budget, data usage, or transaction partners.

Secondly, the trust issue on the action chain formed between intelligent agents.

An intelligent agent may further call upon other agents, models, data, and tools during task execution, forming a continuous A2A delegation. The bias in the judgment of the preceding steps may continue to amplify along the task chain, and each task disassembly and permission transfer may also blur the original authorization boundaries, ultimately leading to greater synthesis bias. Once a problem arises, it is also difficult to determine where or at which stage the problem occurred. Therefore, the intelligent agent economy needs to address not only whether a single intelligent agent is trustworthy, but also whether the entire action chain can be observed, verified, and held accountable.

The principal-agent relationship has extended from between people to between people and machines, and machines to machines, completely changing the foundation of trust building. The traditional way of relying on contract constraints, outcome incentives, supervision mechanisms, and reputation effects to form trust is difficult to apply because intelligent agents do not have the ability to independently assume legal responsibility, nor do they have independent personality and moral consciousness. Their high-speed operation makes real-time supervision even more difficult. Therefore, the trustworthy actions of intelligent agents require a new governance mechanism and rules that cover pre event, during event, and post event.

Identity and authorization mechanisms need to be established in advance. The identity mechanism should clarify who the intelligent agent acts on behalf of, who creates it, and who authorizes it; The natural language intention of humans needs to be transformed into permission rules that intelligent agents can execute. Through certain authorization mechanisms, it is clear which goods and services the intelligent agent can trade, which resources to call, how much budget to use, whether to allow further delegation to other intelligent agents, and which key links must be reconfirmed by humans.

It is necessary to establish transaction rules and process verification mechanisms during the process. The actions of intelligent agents cannot rely solely on model autonomous judgment, but must transform service conditions, delivery standards, acceptance rules, and payment conditions into machine-readable and automatically executable rules. Whether the task is completed as required, whether the data and tools are called as agreed, and whether the payment conditions are met all require the formation of verifiable states and vouchers during the action process.

Afterwards, it is necessary to establish a behavior audit and correction mechanism. When the intelligent agent deviates from the target, exceeds its authority, or triggers abnormal transactions, the system should be able to pause its actions in a timely manner and transfer them to manual processing according to the degree of risk. After a performance dispute or actual loss occurs, it is also necessary to identify the problem link along the task and transaction records, and clarify the responsibilities of the relevant parties.

In the future, the identity system, authorization mechanism, secure payment, and dispute resolution around intelligent agents will not only be the backend capabilities of the market, but also the governance center of the entire proxy network, responsible for transforming economic intentions into executable transactions and supporting the operation of the intelligent agent's trusted action network.

Weakening of traffic logic and rise of trust logic

The changes in transaction costs will not only alter the core issues that the market needs to address, but also reshape the boundaries of enterprises and the way economies of scale are formed. External transactions of intelligent agent systems will shift from topology dominated mode to throughput dominated mode, with coordination costs mainly increasing with task execution rather than the number of connected nodes. The rate of decrease in external transaction costs is much faster than the rate of decrease in internal coordination costs. The result drives the expansion of enterprise scale, closed integration towards more open professional collaboration.

At the same time, as trustworthy actions become the new core market friction, the focus of enterprise competition will gradually shift from competing for traffic entry to establishing connections and trust in the intelligent agent ecosystem.

Firstly, the boundaries of enterprises will shift from closed integration to open calling, and professional division of labor will be further deepened. According to Coase's theory of enterprise, whether a capability is provided internally or acquired externally through the market depends on the comparison between internal organizational costs and external transaction costs.

In the Internet era, in order to reduce the cost of cross subject communication and coordination, enterprises tend to incorporate more businesses and services into their own systems to form large platforms or super apps. Intelligent agents can actively discover, compare, and invoke external capabilities, making models, data, tools, professional knowledge, and business processes identifiable and composable by machines in the market supply.

Enterprises no longer need to independently build all the capabilities required to complete a task, but can dynamically organize external resources around specific goals; The internal capabilities that have their own advantages can also be encapsulated into services for other enterprises and intelligent agents to call upon.

The competitiveness of future enterprises no longer depends solely on how many resources they have internally, but also on their ability to effectively organize external capabilities, including how many agents to orchestrate, how many APIs to call, and how much data and tools to use, to jointly accomplish tasks.

Secondly, the demand entry points are more diverse and the supply is more transferable, significantly weakening the advantages of super apps. In the Internet era, people need to enter specific apps to search and select services, so platforms and enterprises compete around people's attention, dwell time and traffic portals.

In the era of intelligent agents, there is no need for people to enter each platform separately to search for services, but rather for agents to act as agents; Requirements do not necessarily have to be limited to specific apps, but can be initiated from multiple sources such as mobile systems, smart terminals, enterprise office systems, and various professional intelligent agents. Meanwhile, goods and services will gradually form machine-readable expressions, and digital resources can be discovered and called by different intelligent agents.

Traffic may still be important, but it is no longer the only bridge connecting demand and supply. The focus of enterprise management will gradually shift from how to keep users and services within its own system to how to enable more intelligent agents to accurately understand, actively select, and conveniently call upon its products, services, and professional capabilities.

Thirdly, the scale effect will not disappear, but it will shift from traffic scale to intelligent agent ecology. Every time an intelligent agent executes a complex task, it needs to perform model inference, data calling, resource coordination, and result verification. Different industrial scenarios also have significant differences in professional knowledge, business processes, and responsibility requirements.

Therefore, agent services for specific tasks do not necessarily replicate the growth pattern of continuous decline in marginal costs of Internet platforms, and the application and performance layers may present a more professional and vertical development pattern. But in a broader ecological layer, the connections and collaborations between intelligent agents still have significant network effects. The richer the supply of intelligent agents, merchants, and professionals that an ecosystem can connect with, the more types of tasks it can support, and the greater its value to market participants.

Trustworthy actions will become the key to the formation of ecological network effects. The value of an ecosystem depends not only on how many intelligent agents and professional suppliers are connected, but also on whether these entities can collaborate based on common identity, authorization, payment, verification, and responsibility rules.

The more trusted rules are accepted by market entities, the lower the cost of establishing connections and completing transactions between intelligent agents; The addition of more participants will further enrich the supply, expand the scenarios, and enhance the ecological value. The scale effect gradually shifts from the accumulation of user numbers and traffic to the ecological synergy formed by the scale of intelligent agents, professional supply networks, and trusted transaction rules.

The intelligent agent ecosystem should not be a new entry point for a few closed control enterprises, but an open collaborative network composed of multiple demand entry points, multiple types of professional supply, and common trusted rules. Different enterprises can play different roles in it: some are closer to the true intentions of users and enterprises, some have profound industry knowledge and professional supply, and some provide basic capabilities such as identity, payment, verification, and auditing.

Future enterprises may not necessarily become intelligent agent platforms, but they need to find their place in the mobile network. The long-term value will depend on what irreplaceable value the enterprise can create for the entire task chain, and whether this value can be reliably, openly, and scaled up by intelligent agents.

Intelligent agent economy leverages new growth drivers

The expansion of the execution carrier of economic activities from humans to intelligent agents will not only reshape market functions and enterprise boundaries, but also ultimately reflect in macroeconomic growth. According to the neoclassical growth theory and its growth accounting framework (Cobb Douglas production function), economic growth mainly comes from capital accumulation, increase in effective labor supply, and improvement in total factor productivity. The intelligent agent economy simultaneously affects these three aspects and is expected to become an important increment in the future macroeconomy.

Firstly, the demand for AI will expand capital investment and broaden the connotation of productive capital. The growth in the number of intelligent agents and the scale of task execution will continue to increase the demand for computing power, basic models, data centers, and cloud computing, directly driving investment in AI infrastructure. At the same time, intelligent agents can organize dispersed models, data, software tools, and enterprise knowledge around specific production goals, forming production capabilities that can be continuously called upon and reused across scenarios. The past business experience that relied on individuals can also be accumulated into a knowledge base, process rules, and task templates, and transformed into intelligent capital that enterprises can replicate and accumulate.

Secondly, intelligent agents will amplify individual capabilities and expand the effective labor supply in the economy. The labor input in growth theory depends not only on the number of workers and working hours, but also on the quantity, complexity, and value of tasks that can be completed by a unit of labor. After the intelligent agent enters the production process, workers will gradually shift from personally completing each step to setting goals, organizing resources, supervising the process, and judging results. The effective labor of the future will be a combination of human judgment, intelligent agent calling ability, and supervision and correction ability. Individuals and small teams can also use intelligent agents to acquire professional capabilities that were previously only available to large institutions, thereby expanding the scope of economic activities they can participate in.

Thirdly, intelligent agents will reduce innovation costs and enhance total factor productivity. The endogenous growth theory represented by Romer emphasizes that knowledge production, technological innovation, and knowledge spillovers are important sources of long-term growth. Intelligent agents can enter knowledge intensive processes such as research and development, design, programming, and analysis, reducing the cost of information retrieval, scheme comparison, and repeated trial and error; It is also possible to lower the threshold for enterprises to apply new technologies and shorten the cycle from technology generation, adaptation to the formation of real productivity through natural language understanding and tool invocation. Professional knowledge can also be encapsulated as tools, interfaces, or specialized agents, enabling capabilities that were originally concentrated in a few experts and large institutions to enter the production activities of small and medium-sized enterprises and ordinary workers more quickly.

In summary, to measure whether the intelligent agent economy can become a true economic increment, it is not only about how much computing power, models, and intelligent agents are added, but also about whether it can help every enterprise organize resources more efficiently, help every worker create higher value, and especially enable small and medium-sized enterprises to obtain professional capabilities that were difficult to have in the past at lower costs, and better serve more users.

The intelligent agent economy ultimately needs to serve people and life. When technological capabilities are truly transformed into business growth for businesses, improvement in labor capabilities, and improvement in consumer experience, a technological breakthrough can be transformed into sustainable economic growth and broader social welfare.

Agent economy is not a simple extension of Internet economy, but a new direction of future economic growth and a new stage of digital economy evolution. However, technological capabilities do not automatically translate into economic growth, and the model's ability to understand goals and execute tasks does not mean that individuals and businesses can already entrust real transactions to it.

To truly operate the intelligent agent economy, a lot of demand and supply side transformation is needed, establishing industry consensus standards and rules, and forming new commercial infrastructure around identity, authorization, and trusted transactions. Only when technological capabilities, industrial scenarios, and trustworthy mechanisms mature together can intelligent agents transform from smart tools into reliable productivity, and transform technological breakthroughs into solid incremental economic activities.

The experience of relying on traffic entry points, closed ecosystems, and internal integration in the past cannot answer new questions in the era of intelligent agents. Faced with the new trend of intelligent agent economy, we should actively embrace changes and explore new innovative paths in real scenarios.

Enterprises should open up their professional capabilities and explore new service models with more partners. The platform should also shift from managing traffic and expanding scale boundaries to connecting diverse demands, organizing professional supply, and establishing trustworthy transaction rules and protocol open networks.

(Li Zhenhua, Dean of Ant Group Research Institute, Zhang Yun, Senior Researcher of Ant Group Research Institute)

(Author's Way of Numerical Wisdom)