When AI really starts working, what can it bring to the economy?

On September 10th, the 2026 Inclusion · Bund Conference officially opened on the banks of the Huangpu River in Shanghai. The theme of this year's conference is "Co creating AI New Economy".

Compared to the past few years, this is a significant change.

Since the emergence of ChatGPT, discussions on artificial intelligence have long revolved around parameters, computing power, scaling laws, and model capabilities. People care about whose model is stronger, who has more GPUs, and on which benchmark the next generation of models will surpass humans.

But three years later, when big models began to move from chatting to doing things, intelligent agents began to enter real transactions, and robots began to enter the frontline production process, the problem had quietly changed: we no longer only discussed what AI could do, but also began to discuss what AI could create - new transactions, new organizations, new productivity, and new economic increments.

This also constitutes the most noteworthy hidden thread of this year's Bund Conference.

Economists, scientists, and entrepreneurs on stage are still discussing models, intelligent agents, quantum computing, and scientific discoveries, but more and more discussions have fallen on the oldest questions in economics: where does productivity come from? How does innovation turn into growth? How can new technologies create new businesses and jobs? How can technological progress ultimately translate into the prosperity of the entire society?

Artificial intelligence is gradually transforming from a technological variable to an economic variable.

From obedience to action

In the past few years, the most impressive thing that big models have left on ordinary people is that they are becoming more and more "able to speak".

Ask a question that provides an answer; Let it write an email, it generates an email; Give it a material and it will complete the summary. The logic behind it is still that humans issue commands and AI responds to them.

But in 2026, this logic is changing.

Source Code Capital Investment Partner and Foreign Academician of the National Academy of Engineering, Zhang Hongjiang, summarized an evolutionary path at the main forum: Agents are "moving from dialogue to assistants, from passive to active, and from individuals to groups".

The truly significant aspect of this statement is that AI is beginning to possess the ability to take action.

A large model that only answers questions is essentially still an information tool; But when the agent begins to understand tasks, call tools, formulate steps, and complete tasks, it truly enters the economic process that was originally composed of human labor for the first time. Furthermore, when different agents begin to collaborate, exchange information, or even complete transactions, it changes not only human-computer interaction, but also the connectivity of the economic system itself.

This change has already occurred in very specific scenarios.

At the roundtable discussion on "When Agents Become Trading Entities", Ant Group CEO Han Xinyi gave a small example: a week ago, he used the AI health app "Afu" to buy some healthy snacks such as nuts and seaweed, and spent more than 40 yuan from searching for products to placing orders.

Of course, 40 yuan is not considered a huge transaction. It is worth noting that for the first time, an agent took a step from "telling you what to buy" to "buying things back for you".

OPPO Senior Vice President and Chief Product Officer Liu Zuohu also mentioned that the consumption of "Xiaobu" has nearly doubled within a few months. Although the base is still small, users have become accustomed to letting agents help them shop.

Once AI is able to take action, it inevitably begins to touch payments, accounts, goods, and services. In the past, the basic structure of Internet commerce was "people looking for goods", "people clicking buttons" and "people completing payment"; In the future, it may become humans setting goals, agents searching for solutions, comparing prices, calling services, and then completing transactions.

So, a whole set of commercial infrastructure must also change accordingly.

Does the agent have an identity? Who authorized it to spend money? Who is responsible for buying the wrong thing? How to trace a transaction completed by a machine?

Han Xinyi summarized it as authorization, identity, capability assessment, fund security, and traceable auditing. In the traditional Internet, people are familiar with KYC - Know Your Customer; In the era of intelligent agents, a new problem is emerging: KYA, Know Your Agent.

Jorn Lambert, Chief Product Officer of Mastercard, also mentioned the importance of this business trust. Alibaba Group Chief Scientist Zhou Jingren believes that as long as the constraints, secure execution environment, and model boundaries can be clearly defined, the trust problem of intelligent agents is not unsolvable.

As more and more machines are able to autonomously complete tasks, conduct transactions, and form collaborative networks, a true intelligent agent economy may emerge.

And this' ability to act 'has also begun to enter the physical world from the digital world. At the embodied intelligence roundtable on that day, several robot entrepreneurs formed a common judgment: embodied intelligence will not simply replicate the development path of large models. What truly determines whether robots can leave the demo is not just the model ability, but the complete ability composed of data, hardware, system stability, and real-world ROI.

The ultimate answer for robots is not whether they can move, but whether they can stably enter factories, stores, and service scenarios, truly becoming a part of productivity.

From Base to Growth

In the past few years, one of the most important keywords in the AI industry has been "base": larger models, more advanced chips, more data centers, and larger capital expenditures.

But looking back at each technological revolution, what truly changed the world for railways was not the steel rails themselves, but the re movement of goods and population as a result; The value of the Internet does not lay much optical fiber, but grows out of new economic forms such as e-commerce, search, social networking and mobile payment.

Is AI also going through the same process?

Nobel laureate in economics, Philip Agion, provided a direct answer using the theory of economic growth.

According to his and his collaborators' calculations, considering only the tasks in AI automated production of goods and services, it may increase productivity growth by an additional 0.68 percentage points per year in the next decade; If we further consider the promoting effect of AI on "idea production", that is, the innovation process itself, it may increase by about 0.4 percentage points.

In other words, 0.68 percentage points may even be the lower limit.

This may also be the most important layer in understanding the "AI new economy": what is truly worth paying attention to is not how much revenue the AI industry itself has created, but whether the production function of the entire economy has changed as a result.

In the past, expanding output for a company often meant increasing employees, capital, and equipment; In the future, as more and more knowledge work can be entrusted to agents, and the experience, processes, and knowledge within the enterprise can be modeled, the same scale of people and capital may produce completely different orders of magnitude of output.

Zhang Hongjiang believes that in the intelligent agent economy, the definition of enterprise assets itself may also be rewritten. In the past, the core assets of enterprises were talent, customers, brands, and patents; In the future, computing power, proprietary models, high-quality data, agent collaboration systems, and continuously iterative workflows will also become new means of production.

If the Internet reduces the cost of information dissemination and transaction, then agents are further reducing the cost of knowledge labor, coordination and implementation.

An entrepreneur can call on a group of agents to complete programming, design, marketing, customer service, and operations, and a "one person+a group of agents" enterprise is no longer just an imagination. Enterprises may become smaller, but their capabilities may actually increase.

It is precisely here that AI has truly transformed from a cost item for an enterprise to a growth item for the first time.

Of course, not all cutting-edge technologies will immediately become productivity. Lu Chaoyang, Executive Dean of the Shanghai Research Institute of the University of Science and Technology of China, gave a cool down to the hot topic of quantum computing at the main forum: quantum computing is not a super GPU that can "accelerate everything", and there is currently no recognized algorithm that can accelerate the training of large models.

This actually provides another scale for understanding the "AI new economy": what really matters is never how cutting-edge the technology sounds. The greatest value of technological revolution has never been to make existing things faster, but to constantly create supply and demand that did not exist before. Steam engines create railways, electricity creates modern manufacturing, and the Internet creates platform economy. What AI can ultimately create, of course, is still not fully answered today.

But the contours have already begun to appear.

From machines to humans

All technological revolutions will eventually return to a fundamental question: what about people?

In a conversation on the main forum of the day, writer Liu Zhenyun and Dean Ma Yi of the School of Computing and Data Science at the University of Hong Kong, one starting from literature and the other from artificial intelligence, discussed the same question - as machines become smarter, what should humans really guard against?

Liu Zhenyun shared a very interesting experience. There are a large number of videos online today that show his face and imitate his voice to express opinions, and '95% of them are fake'. AI can even continue to "speak for him" through "Chicken Feathers on the Ground" and "Ten Thousand Sentences in One Sentence".

But in his view, imitating is not equivalent to creating.

It can imitate 'Chicken Feathers in One Field' and write 'Goose Feathers in One Field', but for works that I haven't thought of or written yet, AI definitely can't imitate them. ”

Ma Yi summarized this difference as "commonality" and "individuality". Today's big models are able to grasp literary, historical, and scientific knowledge well, but these are more common knowledge accumulated by humans. And everyone's emotions, values, and life experiences are different.

Machines are becoming increasingly adept at mastering commonalities, while humans need to find their own individuality even more.

This also provides another perspective on the most common AI anxiety today.

Faced with the question from students and parents about "what to learn to avoid being replaced by AI", Ma Yi believes that finding a skill that AI will never learn may itself be an outdated approach. Instead of avoiding AI, it's better to learn how to leverage AI and amplify the abilities you already possess.

Liu Zhenyun uses rural mechanization to explain employment anxiety. In the past, agriculture relied heavily on human labor. Later, tractors and harvesters entered rural areas, and one person could grow hundreds of acres of land, but people did not stop working because of this. New productive forces will eventually form new production relations.

AI is not a ferocious beast at all. ”

What is truly alarming may not be that machines are becoming more and more like humans, but that humans are becoming less and less like themselves after using machines.

Liu Zhenyun said that when he goes to Hong Kong, he can certainly ask AI which restaurant is delicious, but he prefers to ask Ma Yi, "When will you treat me to a meal?" AI can recommend restaurants, but it won't actually treat you to a meal.

A joke that points out the boundaries that technology can never bypass.

AI can provide knowledge, but cannot completely replace relationships; It can simulate communication, but it cannot replace the real connection between people. Ma Yi also reminds that if all knowledge acquisition shifts to AI, students may be able to learn more freely, but they may also become more isolated. People ultimately have to enter society and interact with real people.

And in farther places, AI may even help people create new knowledge.

Princeton University professor Wang Mengdi raised another question on that day: How far is AI from truly autonomous discovery?

Today's big models are good at learning the most common and mainstream knowledge in probability distributions, but the truly important scientific discoveries often happen outside the long tail. She and her team are trying to truly integrate AI into the experimental world: proposing hypotheses, calling experimental equipment, observing results, and then re proposing hypotheses.

If the main way for AI to create value today is still to improve the efficiency of existing work, then the truly huge increment in the future is likely to come from the second layer of capabilities - creating new knowledge, new materials, new drugs, new production processes, and industries that do not yet exist today.

This is also the meaning of what Agion called "creative destruction".

Innovation eliminates old technologies, old enterprises, and old jobs on one hand, while constantly creating new enterprises, new demands, and new employment opportunities on the other. AI will accelerate this process.

So, the real question may never be whether AI can replace humans, but whether we can make the speed of creation ultimately exceed the speed of destruction; Can everyone still find their place after machines become increasingly powerful.

This also means that the 'AI new economy' is not just a matter for technology companies. It requires new payment and identity systems, competition policies that adapt to technological changes, an education system that helps people continue to learn, and a new boundary between humans and machines.

Last year at the Bund Conference, we were still searching for the path for AI to move from cutting-edge imagination to the real world. One year later, the problem has changed again.

AI is still a rapidly evolving technology, but it is no longer just the technology itself. It begins to buy things, complete work, enter the enterprise, and also begins to redefine the value of assets, knowledge, and people.

The Huangpu River continues to flow eastward. When every technological revolution truly arrives, people are often already immersed in it, but it is difficult to accurately say which day history began to turn.

But at least at the Bund Conference in 2026, one change is clear enough:

The AI new economy is shifting from a story of technology to a story of economy and people.