Tan Liang (Assistant Dean of Custom Executive Education, Professor of Practice in Management and Business Strategy, Faculty of Business and Economics, The University of Hong Kong; former President of GE University China; former Global Executive President of Alibaba Group University)
In 2026, regardless of its external manifestations, the business model will essentially and comprehensively enter the A2A—Agent to Agent—era. In Tan Liang's view, this is not an iterative change, but a reconstructive one: above the computing power layer, model layer, and platform tool layer, a brand-new business infrastructure layer called the "Agent Marketplace" is emerging. Organizational reconstruction must focus on three main contradictions—the "New Three Dimensions" of AI concentration, talent density, and organizational posture. And at the end of the K-shaped divergence lie two employment states: industry veterans having "no jobs to leave," and simple information movers having "no jobs to take." The future is not just approaching; it is already happening.
Editor's Note: On July 17, the 2026 World Artificial Intelligence Conference (WAIC) and the High-Level Meeting on Global Governance of AI opened in Shanghai, and the World Artificial Intelligence Cooperation Organization was established in Shanghai. On the morning of July 20, the third and final session of the WAIC 2026 series breakfast meetings, hosted by Guancha.cn's Science and Technology Power, titled "AI, Organization, and Humans: Current Trends and Historical Rhymes," was held in Shanghai. The first session, "AI for All: Beyond Silicon Valley, the Global South," explored the base of the pyramid; the second, "The Crucial Battle of World Models," looked up to the peak; and this session returned to the middle of the pyramid—enterprises, organizations, and humans. Tan Liang, Assistant Dean of the Faculty of Business and Economics at The University of Hong Kong; Zhang Lei, Entrepreneur in Residence at InnoFund; Xiao Mafeng, Founder of TTC; and Sun Xi, Associate Professor at Capital University of Economics and Business, shared their insights on the same stage. Zhou Yufang, CSO of Guancha.cn, moderated the event. This article is a transcript of Tan Liang's speech and roundtable discussion, organized and edited by Guancha.cn based on on-site audio recordings. Subheadings were added by the editors.
Below is the full transcript:
Time Will Ultimately Reward the Brave
Today, the theme of my sharing is: In the A2A era, how will organizations be reconstructed?
First, let me share a distinct perspective: In 2026, regardless of its external manifestations, the business model will essentially and comprehensively enter the A2A model. A2A refers to Agent to Agent. In the past, when talking about the AI era, we always spoke abstractly; in 2026, we need to make it concrete—what exactly is the AI era? Specifically, it is A2A.
Previously, many friends were engaged in the B2B model, and some were more consumer-oriented (to C); now, it is entirely to A. Behind this, I would like to share two insights.
First, what does A2A mean for everyone in business? This is a cognitive topic. My view is: it should be a brand-new production paradigm. The impact it brings, I understand in the context of humanity's invention of fire and electricity—fire, electricity, and now AI. Fire liberated human physical strength, and our smallest combat unit moved from the individual to the group; electricity released productive capacity, bringing modern industry; AI brings intelligence, liberating the intelligence of the entire human-machine collaboration.
A New Layer is Growing in the Industry Chain: Agent Marketplace
How should we understand A2A behind the entire industry chain? As we know, the bottom layer is the computing power layer, above it is the model layer, and further above is the platform tool layer. According to previous understanding, the platform tool layer directly faces the enterprise application layer.
However, from last year to this year, a new layer has grown between the platform tools and enterprise applications. According to my own understanding, I call it the Agent Marketplace. On this layer, many companies will develop industry-level agent applications and agent development. I originally thought it would take three years for this layer to emerge, but I didn't expect that while visiting WAIC this time, I had already seen many companies making strides in the Agent Marketplace layer—the actual speed is much faster than I anticipated.
Therefore, the A2A era is not just approaching; it is happening right now.
Behind A2A, everyone needs to have a mental picture. In the past, when we did business, marketing, and sales, we faced humans; in the future A2A model, our marketing, sales, and the series of methods behind them will all face agents. Count how many layers there are: computing power layer, model layer, platform tool layer, above that is the Agent Marketplace, and above that is enterprise applications. What changes will happen in enterprise applications? First, there will be an enterprise-level agent layer inside, called the agent middle platform—you can understand it as the agent OS system; driven by it, each department has its own departmental agent. So enterprise applications are divided into two layers: the agent macro-middle platform and the agent applications of various departments.
What about the Agent Marketplace? It connects upwards to the enterprise's agent middle platform and departmental agent applications, and downwards to various tools of super-large platforms—Alibaba Cloud's Bailian, Microsoft's Copilot, as well as Meta, OpenAI, Anthropic, Palantir, and so on. Behind the super-large platforms are models and computing power. The Agent Marketplace will become the next brand-new business infrastructure in the A2A era.
Overestimating One Year, Underestimating Ten Years; Three Enterprise Forms
Historically, with technological disruption, we all have a tendency: to overestimate the impact within one year and underestimate the impact after ten years. This is the first point.
The second point: in the process of major technological changes bringing major disruptions, the market will quickly differentiate into three enterprise forms.
The first is called the native state, accounting for 1%. The second is called the evolutionary state; evolutionary organizations are "Made for humans with AI," accounting for 20%. The third is called the pilot-observing state, accounting for 80%.
A student told me: Professor, there is a slight problem with your logic, how can it add up to 101? Yes, exactly, it is101. Because that 1% was not born in the old era; it was born right now, a new species of the new era, so it is purely native. And the 100% all came from the old era, an 80/20 split: 20% evolved very quickly, iterating and evolving; 80% are piloting and observing.
Therefore, under the great technological disruption, enterprise forms are divided into three categories: purely native, rapidly evolving, and piloting/observing.
Industry Seeks Stability, Internet Seeks Agility, What About the AI Era?
Let me first share my background: 10 years at GE (General Electric), 7 years at Alibaba, and now in my third year at HKU (The University of Hong Kong). I am very grateful for the opportunity to keep up with the times and personally experience three eras: industry, internet, and AI. Therefore, on the organizational side, I have always had a deep physical sense of the changes.
The organization in the industrial era primarily focused on the stability behind the system: maintaining stability through a hierarchical pyramid structure. The basic elements in the organization are positions, above which are departments, and above departments is the entire business. The core keywords are setting goals, decomposing, aligning, and collaborating.
In the internet era, the most core keyword shifted from the stability behind the system to the agility behind speed—it must be fast, with no fixed form, changing very quickly. Thus, new organizational forms emerged, including the exploration of large middle platforms and small frontends represented by Alibaba, as well as "context over control" mentioned by other companies—providing context without excessive control.
Here comes the question: In the AI era, especially as we comprehensively enter the A2A era this year, how will organizations be reconstructed?
Let me first explain a concept: beyond reconstruction is transformation, and transformation and reconstruction are different. What is transformation? Introducing new elements under the old ideological framework of the old era, that is transformation. What is reconstruction? Only when the paradigm is broken is it called reconstruction.
There are many perspectives and dimensions for organizational reconstruction. Personally, I hope to grasp the principal contradictions together with everyone. I recently proposed that behind organizational reconstruction, there is a "New Three Dimensions" approach: the first dimension is AI concentration, the second is talent density, and the third is organizational posture. These three dimensions will reshape the organization and give rise to two types of organizations.
What is AI concentration? There are two directions here: one is called "AI in all," and the other is called "All in AI," representing very different concentrations.
"AI in all" is the initial stage: it looks lively on the surface. Every department now has AI and is running pilots. It doesn't solve problems but creates new ones. It seems every department in every company has AI, but ultimately your financial statements and your metrics remain unchanged.
What is "All in AI"? It is reconstruction. "AI in all" means you "make the business and the company for humans with AI"—the company is still human-centric, with AI added as an appendage. "All in AI" is "made for AI"—your company is built for AI, built on AI, reshaping the entire business flow and business model based on AI.
Therefore, "AI in all" is the initial stage: deploying systems, deploying tools, everyone learning to use them; "All in AI" is absolutely a top-leader project, top-down, based on cognition, reshaping the enterprise form, and building the enterprise on AI. This is the first dimension, AI concentration.
The Second of the New Three Dimensions: Talent Density—Industry Veterans and Fresh Builders
The second dimension is talent density. I have a distinct perspective: in the A2A era, talent density will ultimately become a pole in reconstructing the organization.
In the future, it will mainly consist of two core roles: one is called industry veterans, and the other is called Fresh AI Builders—native AI builders. These two groups jointly build a dumbbell-shaped architecture: one or two industry veterans leading a group of Fresh Builders, backed by tens of thousands of the "Lobster Army," forming an AI-native team.
The first question worth pondering in talent density: do we have industry veterans in our enterprises? If so, where are they? Who are they? How are they defined? They are absolutely not defined by years of service, performance, or goals. An industry veteran must have a deep understanding of the industry and the sector, possessing that spark-of-lightning deep insight.
In my own work, from GE to Alibaba and then to HKU, what has remained unchanged is cultivating top leaders with an entrepreneurial spirit. My observation is that there are three stages on the path to becoming an industry veteran:
First, it starts with intuition. Intuition means you have a love and a passion for this industry. Second, massive rational training. In the industrial era, they said you need to work for 10 years; in the AI era, with human-machine collaboration, you might achieve the effect of the previous 10 years in three to five years—but these three to five years involve massive training starting from passion. Third, returning to intuition. After returning to intuition, a mountain is still a mountain, and water is still water. If they lack the stage of massive rational training in the middle, their intuition will be exactly the same as their initial intuition, without any change.
Behind the industry veteran is another role: the Fresh AI Builder. Young, hands-on, and learning extremely fast. This year, several major companies have comprehensively reformed their campus recruitment, called "no interviews, only tests": they give you an OpenClaw account, and after 48 hours, you bring what you've built with your own hands to the site, and engineers test it on the spot. This is a brand-new form targeting young people.
How to find these Fresh Builders? I have a self-summary: in the AI era, for a person to keep growing, they rely on three major traits. First, they must have inner drive (self-motivation)—not being demanded by the enterprise, but demanding of oneself. In the industrial era, people were constantly demanded by the enterprise and the department to achieve external goals; in the AI era, facing "innovation out of nothing" and the "trend of omnipotence," people must return to their own inner drive, using AI to ask themselves: What do I want? What do I want to create? Second, behind creation, control your ego, maintain openness, be compatible with the world, open yourself up, and constantly listen, see, and learn. Third, continuous learning ability. In short: in the AI era, it's time to start school.
The Third of the New Three Dimensions: Organizational Posture—From Solid to Liquid
The third dimension is organizational posture.
In the industrial era, organizations were solid. Because they needed to solve for stability: the external environment was stable, so the organization used a solid approach to achieve consistent action and stable structure. Solid organizations have an assumption: the environment is stable, the logic is clear, and work can be planned for a year with three years of stable execution. The core of a solid organization is execution behind stability.
In the AI era, especially the A2A era, organizations need to move towards being liquid. What is a liquid organization? "Be water, my friend." This is the philosophy of water mentioned by Bruce Lee in a Western interview in the 1970s.
What does the philosophy of water represent? I summarize it in 16 Chinese characters, which is the organizational form in the AI era: Ready when summoned, lethal when deployed, victorious when engaged, dispersed when victorious.
In the past, everyone had a fixed position and fixed headcount—the solid organization in the industrial era was called the "Three Fixes": fixing establishment, fixing positions, fixing headcount. In the liquid organization of the AI era, there is no such thing as departments and positions. My extreme personal view is: positions will become increasingly blurred, and what will become very important? The "New Three Items": first, project-driven; second, projects facing scenarios; third, task flows embedded within scenarios. So the new three fixes are: fixing scenarios, fixing projects, fixing tasks. The old three fixes, the new three fixes. Organizational posture also moves from solid to liquid.
Under the New Three Dimensions, Two Types of Organizations Emerge
To summarize briefly: two major directions for AI concentration, two major roles for talent density, and two states for organizational posture. These three perspectives and three poles will ultimately give birth to two new types of organizations.
The first type is the evolutionary organization. Those of us coming from the old era need to evolve quickly. What does its new three dimensions look like? In terms of AI concentration, it can start from the status quo, quickly pass through "AI in all," and sprint towards "All in AI." In terms of talent density, it has industry veterans, but not many Fresh Builders, facing the need to transform a large number of frontline employees—so it needs to create an atmosphere to inspire a large number of frontline employees, giving them self-drive to move towards becoming the industry veterans in their minds. In terms of organizational posture, it is mostly still solid, but it must quickly pilot liquid organizational forms in local areas.
The second type is the AI-native organization. In terms of AI concentration, it is absolutely "All in AI"—it is inherently made for AI. In terms of talent density, first, it has very solid industry veterans: industry veterans paired with chief scientists; second, it is 20% industry veterans multiplied by 80% Fresh AI Builders, multiplied again by the agent team.
What is the agent team? It is an amplifier. Equipped with an agent team, it can multiply your enterprise's talent density by ten, a hundred, a thousand—it will ultimately depend on the multiplier effect of the three elements. Multiplying by ten is when the agent team is just piloting and hasn't deeply reformed the workflow; multiplying by a hundred is when the agent team has reformed the workflow and the business model is being reshaped; multiplying by a thousand is when it has run through and is rapidly replicated. In terms of organizational posture, the AI-native organization is purely liquid, following the new three fixes, constantly moving towards being ready when summoned, lethal when deployed, victorious when engaged, and dispersed when victorious.
Today's opening sharing consists of three points: what is A2A; what are the new three dimensions; and under the new three dimensions, what do evolutionary and native organizations look like.
"No Jobs to Leave" and "No Jobs to Take"
After listening to the sharing of the other three guests, I have a strong feeling: this is truly a paradigm-reconstructing great change. In our daily lives, we see a lot of iterative changes, but this is a reconstructive change. Several guests just mentioned a letter multiple times—K, K-shaped divergence. When I thought about this K-shaped divergence some time ago, I felt there might be a kind of cruelty: in the business models of the future A2A era, there is a very critical word called "needed or not."
Many times, you are not needed—it's not about whether you work hard or not, it's that you are not needed. Under this circumstance, the K-shaped divergence upwards and downwards will extend into two employment states.
One is called "no jobs to leave." For example, the industry veterans mentioned earlier, as Professor Sun said, they decide where the "seat" should be—this ability is highly differentiated and particularly valuable. Such people truly have "no jobs to leave": it doesn't mean they sit at their workstations 7x24 hours, but their judgment in their brains is replicated by agents into a capability asset that operates 7x24 hours non-stop.
The other is called "no jobs to take." If we are only doing pure, simple transportation and pure, simple information processing, this capability will be 100% replaced by AI agents in the future—then it will not be needed.
Let me give an example. Didn't a new layer grow in the A2A era, called the Agent Marketplace? The fastest to react to this was a friend of mine, who used to work in investment at an investment company. Basically a month later, he came out to start a business—making agents, producing agents, and enterprise deployment. He led a small team, and their first client was a company doing ISO 9000 certification: 50 people, doing it for decades, with highly standardized processes. They used two weeks to "distill" this company. When talking to the founder, the founder didn't quite understand at first; until they deployed the product features, the founder immediately understood and said one sentence that day: "I think the most wise decision I made this year is that I will retire in the second half of the year."
Because according to this model, within two weeks, all the standardized processes, capabilities, and skills of this company were distilled—in the future, an agent company, a small team, can fully take over a standardized, process-driven business. So behind it is not simply "some people have no jobs to leave, some people have no jobs to take," but rather: some capabilities cannot be dismounted, and some capabilities will not be needed.
A2A is Not Just Approaching, It is Already Happening
The moderator asked if there are practical examples. Yes, they are already realized by others.
The first example is Alibaba.com. This is an old business of Alibaba since 1999, originally doing information matching, a bit like an online yellow pages, later continuously patching to solve payment, trust, and supply chain issues. Doing foreign trade is very closely linked to the US, and the latest changes in Silicon Valley can be captured in time. They started internal agent R&D in 2024, and the first move was to build four agents: risk control, marketing, compliance, and customer service.
Last year, they made a major prediction: the future will move from B2B to Agent to Agent. The first decision made based on this direction was "four in one"—don't grow more agents later, just one agent. They created a new type of agent application called Accio. Accio immediately changed the business model.
As we know, from the industrial era to the internet era, when looking at enterprises, we still looked at "muscle"—revenue and profit. But in the AI era, the most AI metric for an enterprise is ARR (Annual Recurring Revenue). "Because the AI-native business model essentially sells not products, but continuously operating intelligence; the deeper the intelligence operates, the higher the ARR." The original business model of Alibaba.com was membership-based: annual membership fees, tiered memberships, and extra fees during big promotions. After fully switching to Accio, it moved from membership to subscription-based: different tiers, with the underlying layer entirely billed based on token consumption. Look at how huge this model change is.
In the past, when you came to the website, it was buyer-seller interaction; now, Accio is first of all no longer an APP, it is an agent. You raise a demand, and the agent automatically digests, automatically processes, automatically issues orders, and automatically helps you land it. And very soon, they have found: some buyers and sellers are no longer humans directly facing Accio—large buyers already have their own agent teams. So I say, the entrepreneurs who came to me at the end of last year have changed from "Professor, guide us in AI deployment" to "Professor, we want to do agent deployment. Scaled buyers and sellers are all building their own agent teams. Accio is the platform's agent, and both sides have their own agents—now it is agent to agent. It used to be API docking, now it is protocol docking. This example has already happened, and they are tracking the ARR metric every month.
The second example is in Hong Kong. The pillar industry of Hong Kong is finance, and behind finance is a large batch of industry chains called professional services—tax, legal, compliance, branding. Now, many small and medium-sized companies in these professional services are heavily agent-ifying. Last time I talked about this model in an EMBA class, many classmates who started their own businesses in professional services strongly agreed and are also rapidly agent-ifying. Think about it, the future financial industry chain in Hong Kong: in the front there are C-end and B-end, behind them are bank agents, and behind the banks are a large number of professional service agents; small and medium clients and high-net-worth clients will also have their own agents in the future—many family offices are now building their own family office vertical models to help high-net-worth individuals make decisions and conduct professional analysis in more complex scenarios.
So the shock to me is: all this is not just approaching, it is already happening.
To the Young: After Two 70 Years, Return to Being Human
Why discuss this topic in 2026? Because of the significance of two "70 years."
On one side, after 70 years of development, artificial intelligence has reached every C-end individual globally. On the other side, over these 70 years post-WWII, humanity has experienced the maturation of the industrial era, the explosion of the internet era, and then the great social division of labor in the AI era. What impact have these 70 years had on people? After 70 years of division of labor, achieving large-scale social efficiency through processes, everyone has been put into a link in the process. After 70 years, it brings a very important impact: we innately assume that person equals position, and position equals person.
You will find that in the past few years, when asking most people to introduce themselves, it is basically "I am in what position at what company"—rarely do you encounter someone saying "who I am, what my name is, why I am called this name, what my ancestors were." This is not convenient to talk about much in large public occasions, but I feel these 70 years are a process of "alienation": making the person themselves slowly shrink onto the position.
After 70 years, the two sides of a coin meet: one side is artificial intelligence, solving the problems of robots and embodiment; the other side is biotechnology, solving the problems of human longevity and gene recombination. When artificial intelligence meets humanity, the first thing it replaces is what we are best at—the standard, high-frequency work in positions.
After replacement, the biggest impact on us is the loss of value: when all the work I can do is done by AI, then what do I do? Behind "what do I do" is a huge, huge anxiety vacuum—our mindset is person equals position, position equals person, and AI has arrived, creating something out of nothing and being omnipotent.
So it is very meaningful to discuss this topic at this time. Why is most people's reaction to AI anxiety and fear? It is precisely because of these 70 years—if placed 70 years ago, it might be different.
How to break this situation? First, do not treat AI as an opponent. AI is our helper, it helps us return to being human. Return to which aspects? First is aesthetics, second is the most precious inner curiosity of humans, and third is judgment.
Many students ask me: Professor, I agree, but how to do it? Some of the students I am currently leading were born in 2003. My advice to them is:
First, the inner drive of self-motivation. In the future, with technological development, we might have a "100-year life"—the book "The 100-Year Life" wrote about it, and I think it's not a dream. A long career life, no matter how long the cycle, ultimately goes through three stages: first, rapid trial and error; then anchoring in a certain field, becoming a person with a voice in this field; and finally using your own success to help newcomers. If you just entered the workplace, you need to excavate your inner drive—not catching the demands others have for you, but asking yourself: What do I want to do?
Last time, a reporter interviewed Messi before the final, asking him how he viewed the fierce competition during the game. Messi said: We actually don't pay too much attention to the score, we just enjoy the game, we have enjoyed playing football since we were kids. This paragraph touched me greatly. Now when I ask my students, I feel everyone rarely has a sense of enjoyment—because of the 70 years of great social division of labor, everyone has been stuffed into processes, stuffed into big factories, stuffed into large organizations, stuffed into systems, making it hard to find that sense of enjoyment.
The inventor of OpenClaw is Peter Steinberger. If you ask him why he designed OpenClaw back then, he just said two words: Have fun, don't be bored. So, first find what things can make you enjoy and be inspired—this is a starting point.
Second, maintain full openness. With the great technological change, all the entrepreneurs and executives I meet have an unchanging sigh: AI is evolving too fast. C-end products have 2.5 versions a day, B-end products iterate and evolve every 90 days. If a person remains closed, they will quickly fall behind. Maintain openness to see best practices; in the process of seeing best practices, stimulate your own curiosity.
Third, learn towards the truth. Is reading a book learning? Not necessarily. "Learning must ultimately be transformed into verifiable decisions—seeing or hearing anything, you must ask yourself: if I must make a decision based on this cognition tomorrow, how would I choose? If I choose wrong, can I iterate within two weeks?"
In short: in the AI era, it's time to start school. Facing a longer life in the future is a marathon, don't exhaust your energy in a certain stage—there must be a long-term plan. "Therefore, in the future, there will be only two employment states—'no jobs to leave' and 'no jobs to take.' Which one do you want to become? The answer begins to be written from the moment you decide 'it's time to start school.'"