AI's Next Act: From Chatting to Doing

What happens when AI stops just talking and starts doing? That was the question hanging over the Huangpu River on September 10, as the 2026 Bund Conference kicked off with a theme that felt less like tech hype and more like an economic wake-up call: "Creating a New AI Economy."

For years, we've been obsessed with parameters, computing power, and whose model could pass the latest benchmark. But something has quietly shifted. Large language models are moving from chat windows into real transactions, agents are beginning to collaborate, and robots are inching onto factory floors. The question is no longer what can AI do? but what can AI create?

From Obedience to Action

Ask a chatbot to write an email, and it will. But in 2026, AI is learning to act on its own. Zhang Hongjiang, an academician of the U.S. National Academy of Engineering, described the evolution: agents are moving from conversations to assistants, from passive to active, from individuals to groups.

That shift matters more than it sounds. An AI that only answers questions is still just an information tool. But an agent that understands tasks, calls tools, and plans steps is stepping into the economic process itself. When agents start collaborating—exchanging information, even completing transactions—they change how the economic system connects.

Ant Group CEO Han Xinyi shared a small but telling example: last week, he used the AI health app "Afu" to buy nuts and seaweed snacks for just over 40 yuan. Not a huge purchase, but a milestone. An agent has moved from telling you what to buy to buying it for you. OPPO's Chief Product Officer Liu Zuohu noted that similar consumption via "Xiaobu" has nearly doubled in months.

Once AI can act, it touches payments, accounts, goods, and services. The old internet commerce model—people find products, click, pay—may become: people set goals, agents find solutions, compare prices, and complete transactions. But that raises new questions: Does an agent have an identity? Who authorizes it to spend? Who's responsible if it buys the wrong thing? Han Xinyi summed these up as authorization, identity, capability assessment, financial security, and traceable auditing. In the agent era, KYA—Know Your Agent—is the new KYC.

From Foundation to Growth

For years, the AI industry has been building "foundations": bigger models, more chips, larger data centers. But past technological revolutions didn't change the world through rails or fiber alone—they changed it through new economic forms. Nobel laureate Philippe Aghion offered a direct answer: if AI automates production tasks, productivity growth could rise by about 0.68 percentage points per year over the next decade. Factor in AI's boost to innovation, and you can add another 0.4 points. That 0.68 might even be a lower bound.

This is the heart of the "AI New Economy": it's not about how much revenue AI generates, but whether the entire economy's production function changes. When knowledge work can be handed to agents, a startup can call on a group of agents for programming, design, marketing, and operations. "One person + a group of agents" isn't just a slogan. Companies may get smaller—but their capabilities could grow larger.

Of course, not every shiny technology becomes productivity overnight. Lu Chaoyang from the Shanghai Institute of the University of Science and Technology of China poured cold water on quantum computing: it's not a super GPU that accelerates everything. The real value of a tech revolution isn't doing existing things faster—it's creating supply and demand that never existed before.

From Machine to Human

At the end of every tech revolution, we return to a fundamental question: what happens to people? Writer Liu Zhenyun and Professor Ma Yi of the University of Hong Kong tackled this head-on. Liu shared that countless videos online feature his face and voice, "95% of which are fake." AI can even write a "sequel" to his works. But imitation isn't creation. "It can imitate 'One Heap of Chickens' to write 'One Heap of Geese,' but I haven't thought of or written a work that AI can't imitate."

Ma Yi framed it as commonality vs. uniqueness. Today's models master common knowledge well, but everyone's joys, sorrows, and life experiences differ. As machines get better at commonalities, we need to find our own uniqueness more than ever.

So what should students learn to avoid being replaced? Ma Yi thinks that's the wrong question. Instead of avoiding AI, learn to use it to amplify your abilities. Liu Zhenyun compared it to agricultural mechanization: tractors let one person farm hundreds of acres, but people didn't stop working. New productivity forms new production relations. "AI is definitely not a flood or a monster."

What's truly worth worrying about isn't machines becoming more like humans, but humans becoming less like themselves after using machines. Liu joked that he could ask AI for restaurant recommendations in Hong Kong, but he'd rather ask Ma Yi: "When will you treat me to dinner?" AI can recommend a restaurant, but it can't invite you to dinner.

And further out, AI might help us create new knowledge. Princeton's Wang Mengdi asked: how far is AI from true autonomous discovery? Today's models excel at mainstream knowledge, but breakthroughs often happen in the long tail. Her team is trying to let AI propose hypotheses, call experimental equipment, observe results, and re-propose. If today's AI mainly improves efficiency, tomorrow's huge increment may come from creating new materials, new drugs, and industries that don't exist yet.

Key Points

• Agents are becoming economic actors: They're moving from answering questions to completing transactions, forcing new infrastructure like KYA (Know Your Agent).
• Productivity gains are measurable: Nobel laureate Philippe Aghion estimates AI could add 0.68–1.08 percentage points to annual productivity growth over the next decade.
• Companies may shrink but grow stronger: "One person + a group of agents" could redefine corporate assets and capabilities.
• Human uniqueness matters more than ever: As AI masters common knowledge, our individual experiences and relationships become our edge.
• AI's biggest potential: Not just doing existing work faster, but creating entirely new knowledge and industries.