The Tencent Research Institute and the University of Hong Kong Faculty of Law jointly host the AI & Society Forum 2026, themed "Living with AI." The roundtable brings together frontline practitioners such as Hu Yichuan, CTO of Yonyou Technology, Wang Ruoyu, co-founder of Workstream, and Wang Shengjie, product lead of Tencent WorkBody, to discuss AI-driven organizational transformation. Participants believe that AI transformation is a CEO-led initiative—success requires direct involvement from the top executive. Team sizes are shrinking significantly, from 20–30 people down to 3–5. Reducing collaboration numbers and aligning understanding represent the greatest challenges. While super individuals are more productive, they are also busier, as they bear the responsibility of influencing others and reshaping workflows. In the AI era, vision is more important than ever; AI should serve as an amplifier of human capability, not merely an accelerator.
Author and source: Tencent Research Institute
Are we ready for the era of coexisting with AI?
As artificial intelligence reshapes intelligence, productivity, and even social structures at an exponential pace, are we ready for the era of coexisting with AI?
In May 2026, the Tencent Research Institute and the Faculty of Law at the University of Hong Kong jointly hosted the AI & Society (AI&S) Forum 2026. The forum centered on the theme “Living with AI,” spanning the two cities of Hong Kong and Shenzhen: in Hong Kong, we explored “The Boundaries of AI,” examining the collision of ideas and industries; in Shenzhen, we experienced technology’s return to human scale through “unplugged” and “fully plugged” educational settings.
In this series of discussions at the forum, leading scholars, industry pioneers, and frontline practitioners came together to cut through the fog of technology and explore AI’s profound impact on society, the economy, truth, and life.
Below is the roundtable discussion on "Frontline Practices in AI-Driven Organizational Transformation," taking you directly into the frontlines of AI transformation.
Host:
- Yuan Xiaohui (Deputy Director and Senior Expert, Tencent Research Institute)
Guest:
- Hu Yichuan (CTO of Laiye Technology)
- Wang Ruoyu (Co-founder of Workstream)
- Wang Shengjie (Product Lead for Tencent WorkBody)
- Edit and organize:
- Dou Meilei, Senior Researcher at Tencent Institute
[Key Points]
1. AI transformation is a CEO-led initiative. If leadership doesn’t understand AI, the few who do will be overwhelmed with work, inefficient departments will undermine the achievements of efficient ones, and partial transformations will be stifled at the organizational level. Only when the top leader takes direct action can true organizational change be driven.
2. Three Levels of AI Transformation: The first level is AI products for users—great products are luxuries, rare and hard to come by; the second level is adding an AI layer to products to simplify interactions and enhance user experience; the third level is fully transforming the organization into an AI-native workplace. The latter two are not optional—they are matters of survival.
3. The super individual hasn’t become more relaxed despite greater productivity; the "time freed up" doesn’t turn into leisure—because they still carry a mission within the organization: to influence those around them and reshape workflows within the team.
4. Reducing the number of collaborators is the hardest part—aligning understanding is the real challenge. Minimize the number of participants while still ensuring all key points are connected. The issue isn’t just communication overhead—it’s the more dangerous “false alignment”: everyone nodding enthusiastically in the meeting, then doing completely the opposite afterward.
5. Vision becomes even more critical in the AI era. Your only capability is to see 30 years into the future earlier than others and directly transform it with the productivity of three people multiplied a hundredfold.
6. AI should not merely be an accelerator; it should be an amplifier of everyone’s capabilities. There are countless unresolved problems and unmet needs in this world—needs are infinite. The key is not who gets replaced, but whether everyone can access AI and share in the benefits of technology.
Discuss the full text
Yuan Xiaohui: Welcome warmly to our three frontline practitioners! Our earlier report, “From Super Individuals to Super Teams,” was just a starting point, and we’re truly honored to have all three of you here—we’re eager to hear insights from those on the front lines of the industry. Could each of you briefly introduce your company: what kind of business you’re in, the size of your team, and your current state of collaboration with AI?
Hu Yichuan: Hello everyone, I’m honored to be here to share and exchange ideas with you. I’m Hu Yichuan, co-founder of Yedaa Technology. Yedaa is a company specializing in AI digital employees, primarily serving enterprise clients by providing AI-driven digital staff for mid- and back-office knowledge workers in areas such as finance, customer service, and IT. In simple terms, we offer a software solution that automates repetitive tasks in their daily work, freeing up their time and energy to focus on more valuable and creative endeavors.
We’re a team of around 200 people, with roughly 50 focused primarily on AI software development. Over the past two years, our team has undergone tremendous change. I clearly remember: about two years ago, everyone was using GitHub Copilot; one and a half years ago, everyone switched to Cursor; one year ago, everyone adopted Claude Code; and today, everyone is simultaneously using multiple instances of Claude Code and multiple Codex instances. What I want to share is that while our team size has barely changed over the past two years, the number of product lines and the speed of product iterations have increased by at least fivefold.
Yuan Xiaohui: Everyone will also be busier, right?
Hu Yichuan: Yes, everyone is indeed busier. As mentioned earlier, while super individuals are indeed more productive, they are not becoming any less busy—on the contrary, they are even busier now, because within their organizations, they carry the mission of influencing those around them and reshaping internal workflows.
Yuan Xiaohui: Let’s set aside the question of whether we should be busier for now and hear what the next person has to say—Ruoyu?
Wang Ruoyu: Thank you. My name is Wang Ruoyu, and our company is called Workstream—perhaps not well known among friends in China. We address a range of issues for blue-collar workers in North America, from recruitment and onboarding to time tracking and payroll. In the past, we raised approximately $100 million in Silicon Valley and are currently in the Series B stage.
Why does this issue exist? Because white-collar workers don’t actually face this problem—salaries are fixed; many people’s wages at Tencent haven’t changed in three years, right? But blue-collar workers are different, especially in the U.S., where the system is centered around hourly workers. Earlier, a professor mentioned Trump, and his most significant achievement was making tips tax-free in the U.S.—that was Trump’s core accomplishment.
Because tips are tax-free, low-income individuals saw a significant increase in earnings, but this also led to extremely high variability in blue-collar incomes. You can imagine that the number of hours someone works each day is inconsistent. And unlike in China—where, for example, if you run a barbecue restaurant, you might earn 3,000 yuan per month, with meals and accommodation provided, and you’re responsible for everything: taking payments, cutting meat, serving dishes—all on your own.
But in the U.S., it’s different—each position, from cashier to barista, has a different hourly rate, making the data extremely complex. Moreover, blue-collar jobs and their management are a very low-tech industry, so we’re trying to address these challenges using SaaS combined with AI.
Regarding AI, internally we believe there are three levels where it can be applied. The first level is building a product面向用户—this is luxury—because if you can create a product directly for customers, that’s actually quite luxurious, as very few products in the world truly achieve a strong product-to-customer fit. But most companies can at least do two things—the second level: add an AI layer to your product, either to improve efficiency for users, simplify the UI/UX, add a sense of technological sophistication, or enhance your company’s narrative—this layer is essential to implement.
The final layer is what Yichuan mentioned—that every company should do: transform your organization. This is our main topic today: how to turn your company into an AI-native workplace. For example, in the past, you might have needed a CFO accompanied by a team of finance staff manually calculating countless accounts—especially today, if you’re a company operating internationally with entities in China, Singapore, and the U.S., accounting becomes extremely complex. Can AI automate these tasks? Can internal coding be automated? What about product managers? In the past, their core task was writing endless documentation—but that shouldn’t be the case anymore. Today, our product managers do just one thing: engage deeply with users, feed those conversations into AI, and let the AI automatically generate GitHub issues or to-do lists, analyze priorities, and then let engineering teams take over. The entire workflow has undergone a qualitative transformation.
Overall, these three layers represent what we see as actionable opportunities: the product layer is luxury—it’s desirable but not essential; the second and third layers, the AI layer and AI-native work, are both mandatory.
Wang Shengjie: Hello everyone, my name is Wang Shengjie, Jason—you can also call me Jason. I’m from Tencent and currently leading the WorkBuddy product. Regarding AI practices, I’ll start by sharing my own story.
I previously worked on CodeBuddy—an AI programming assistant embedded directly within IDEs. Many of you may have used VS Code or other IDEs; that’s where it was integrated. My story goes like this: I always wanted to use AI not just to help me write code, but to assist me with even more tasks. After writing code, I realized AI could do far more than just help with coding—it could help with a wide range of activities. So I began wondering: could AI help everyone, not just programmers?
So we decided to create a product using Vibe Coding for everyday, non-technical users. At the beginning of this year, I built a web version based on CodeBuddy called CodeBuddy Work. Later, Dr. Xiaohui helped us rename it to WorkBuddy—and thus, WorkBuddy version 0.1 officially launched on January 15th. That’s our story. Our team has also been growing steadily.
We didn’t expect WorkBuddy to be so enthusiastically embraced both internally and externally—this tool truly resonated with users and has high retention, being used daily. Leveraging the viral phenomenon of the "lobster," we integrated the claw functionality and quickly launched a mini-program version—the first dual-formatted solution in China, allowing users to instantly access their mini-program anytime. Today, we’ve also released the APP version, so users worldwide will soon be able to search for and use WorkBuddy.
Yuan Xiaohui: Since he comes from a product management background, some of the terms he just mentioned—like Vibe Coding and IDE—might sound unfamiliar to you. But to put it simply, what is this all about? Because AI has such strong execution capabilities and can code, you can now use natural language to instruct your computer to program and fulfill your requests. For example, your request could be to create a PowerPoint presentation—the very PPT presented earlier in the report was entirely generated by AI after we described it in natural language. This capability boosts efficiency for programmers, but for non-programmers, it represents an entirely new way of working. If you haven’t encountered this yet, you definitely will soon—no matter what kind of product you use, you’ll experience it.
Last year, we released a report on our AI transformation, outlining three levels of enterprise transformation: At the top is business transformation—improving efficiency, attracting more customers, and enhancing R&D quality, as previously mentioned. The middle layer is organizational change—what kind of organization can effectively support AI as a productivity tool? The third layer is mindset transformation—how should we fundamentally view the relationship between humans and AI? This determines both our organization and our business.
My second question is about AI-driven organizational change—your team is roughly 100 to 200 people strong. What have you done so far to adapt to this entirely new productivity tool? Since everyone is now doing not just one thing but potentially many things at once, how should the team divide responsibilities and collaborate? Do we still need so many meetings? Let’s start with Yichuan.
Hu Yichuan: I think the most significant change is that our teams have become much smaller. In the past, a single product line could involve anywhere from twenty to thirty people, including product managers, designers, and front- and back-end engineers; even the smallest teams had around ten to twenty members. Today, each of our product lines consists of fewer than ten people, and many new products built from scratch require only three to five individuals. One reason is that individual productivity has increased dramatically; the other is that by reducing team size, we’ve minimized interpersonal communication overhead. We found that before AI, communication between people consumed the most time and energy.
Yuan Xiaohui: Many people attend meetings during work hours and do the actual work after hours.
Hu Yichuan: Yes, so today I realized that smaller teams are more efficient—they have less communication overhead, perhaps just three people meeting for half an hour each morning to discuss and decide on the most important tasks, then going their separate ways to get things done, letting AI handle the rest. That’s the biggest change.
Another change is the relationship between humans and AI. At the beginning of this year, I set a requirement for the entire product and R&D team: everyone using AI must go through three stages in sequence over the coming months. The first stage is for agents to become your default work entry point—tools like WorkBuddy and Codex should be your default starting point, rather than having to open an IDE or a PowerPoint separately.
In the second stage, building on the first stage—where one person used one agent—now one person can use multiple agents. This is easy to understand: while one agent is working, your time is freed up to do other things. However, this introduces a bottleneck: colleagues may open four, six, or even eight Claude Code windows, but soon realize their attention is stretched too thin, as they constantly need to switch from one window to the next, and then to the next.
So the third stage is designing new workflows that decouple human and AI work—ideally, AI should operate after humans have finished their day. This is what we’re striving for today—though we haven’t fully achieved it yet. Recently, we’ve been tracking the number of tokens consumed by each team and individual, and we’ve seen that some colleagues can have their agents consume billions of tokens overnight after they’ve clocked out.
Yuan Xiaohui: How much is billions of tokens worth?
Hu Yichuan: We all purchased the $200 Coding Plan; if we were to convert it into API usage, I estimate it would cost hundreds of dollars per day.
Yuan Xiaohui: That’s a great answer. We’ve recently researched token consumption across many companies, with monthly usage ranging from a few hundred to several thousand, and even tens of thousands of dollars. The most extreme case was an OpenAI researcher who consumed 217 billion tokens in just one week—equivalent to several thousand dollars.
Hu Yichuan: In the millions.
Yuan Xiaohui: Yes, Peter Steinberger, the founder of OpenClaw, consumed the most—he spent over $1 million last month. Think about it: one person using that many tokens would naturally raise questions: “What did you do with all these tokens?” He didn’t disclose many details, but he said it was still far more cost-effective than hiring more high-salary engineers.
Of course, there are many issues here—employment concerns, job displacement, whether to use humans or agents, and so on. But purely from the perspective of token consumption, it has already brought about significant productivity innovation. Just now, Yichuan also mentioned the overall state of the organization, including how to invoke agents, how to use multi-agent systems, how to protect your attention during this process, and even how to enable collaboration between teams and multi-agent systems—so far, we haven’t seen particularly effective productivity tools. Let’s hear Jason’s thoughts next—whether there will be tools tailored for super teams, which is also essential. Ruyu, how is your team readjusting to build an organization for the AI era?
Wang Ruoyu: I think there are two levels—the tactical level and the strategic level.
How can we improve efficiency at the tactical level? Just now, Yichuan mentioned one point: our internal first principle is that, when doing something, while ensuring all the dots are connected, we should minimize the number of people involved. It’s a bit convoluted to say, but here’s what it means—imagine you’re trying to accomplish anything; there are always several key elements that need to be linked together. For example, building a tech company requires technology, product, marketing, sales, and operations—there are several dots. In the past, you might have needed five people or five departments to collaborate, because each area had only one expert, or the tasks were too fragmented and involved a lot of execution work. But today, because manufacturing and execution have become cheap and fast, these five tasks are best handled by one person. We don’t always achieve this, but if you can get it done by two or three people instead of five, it’s dramatically faster.
It’s simple: when five people are in a meeting room, aligning everyone’s thinking alone wastes a lot of time. And there’s a toxic culture in many offices: one person who seems very knowledgeable speaks, and everyone else nods frantically—not because they understand, but because the atmosphere naturally prompts nodding. The speaker, having a higher level of understanding, genuinely believes everyone gets it and says, “Alright, let’s go ahead and do it”—only for everything to fall apart. This is the biggest bottleneck in management: you think everyone understands, but they actually don’t.
So aligning understanding is the most difficult part. Why do we reduce the number of people in a meeting? It’s not about how many employees the company lays off, but rather, on every matter, we aim to minimize the number of participants, ensuring that everyone involved can cover all necessary points, and ultimately, these points can be connected—then the matter is resolved. This is a tactical approach.
On a strategic level, there’s a deeper pitfall: many believe that hiring a great CTO or someone who knows AI to lead the team is enough—but that’s not the case.
The core bottleneck in an AI organization’s transformation often lies with the top leader—I call this an "owner-driven initiative." In English, you might say, "It’s the responsibility of the owner of the work." If you, as the top leader, don’t actively drive this transformation, problems will arise. Let me give you two examples.
Here’s a smaller example: In a technical team, suppose your manager doesn’t understand AI—say, you’re a seasoned veteran from an old-school software company—but you have a few exceptionally skilled engineers who are driving the AI transformation. The result? Those few individuals end up doing the work of a hundred people, while many others do very little. You keep pushing more tasks onto those two or three people, and because you don’t understand the technology yourself, they’re skilled at managing their relationship with you—and you’re reluctant to let them go. The outcome? The people who get things done end up doing even more, while those who don’t contribute sit back and still get paid. This is an inevitable problem in technical teams. Only when you yourself are deeply proficient in the subject can you accurately identify who truly adds value and who doesn’t.
Pulling it to a higher dimension reveals a deeper issue.
For example, Yi Chuan, as CTO, says he wants to drive an AI-driven organizational transformation, making 200 people as efficient as 2,000. Then the CEO says, "Great, let’s lay people off," reducing the team from 200 to 20 people doing the same work—impressive, right? But what if the CEO doesn’t understand AI? The $180 million saved from laying off those 180 people will end up flowing to other teams that haven’t completed their AI transformation. These teams are already bloated and inefficient; once they absorb the budget, they’ll generate even more false and erroneous demands. Meanwhile, here, the team keeps shrinking, and the remaining capable employees become increasingly overburdened.
These are two classic examples—when the transformation is limited to a team level rather than the entire company, and the top leader doesn’t personally drive it, the very people who get things done end up being crushed by the transformation.
Yuan Xiaohui: That was very vividly put! Indeed—many of the cases in today’s report were driven directly by top executives themselves, including President Si Xiao’s remarks this morning—he is also using AI tools intensively. He noticed a significant shift in his workflow: previously, he might have needed others to help complete certain tasks, but now he can simply direct AI agents to do them himself.
Wang Ruoyu: Yes, because a very practical point is: if you’re in a corporate environment, or even in a university—this applies to HKU as well—you may notice that some professors are AI-native, while others are not. Those who are AI-native have seen their productivity increase by a hundredfold. Those who aren’t are feeling tremendous pressure.
So when the top leader of an organization lacks this understanding themselves, they naturally hear a lot of negative remarks from those who haven’t yet transitioned—these people will claim that others are using AI for harmful purposes, that their speech and outputs haven’t been moderated, and may be non-compliant, filling their ears with nonsense to push them out.
If the top leader of the entire organization doesn’t act as the “catfish”—the so-called catfish effect—then those on the front lines trying to be the catfish will fall behind. This is a tragic aspect of human nature. To summarize: on a tactical level, always strive to minimize the number of people trying to connect all the dots; on a strategic level, if the top leader isn’t fully committed, take a moment to assess your position—don’t become cannon fodder. That’s my advice.
Yuan Xiaohui: Most of the audience here today are business owners, along with many students. From this perspective, the key insight for individuals is “connecting the dots”—you should broaden your skill set to take end-to-end ownership of tasks. It’s not enough to just be strong in your core expertise; you need greater breadth alongside depth—the so-called “T-shaped” talent, with wider coverage across disciplines.
Today at lunch, I spoke with a classmate who has a legal background, and he also noticed this issue: while it’s important to specialize deeply in one direction, you must also broaden your scope—and even adopt an entrepreneur’s mindset, understanding what value you’re truly delivering—this is even more important.
Alright, let’s move on to this—how does your team currently collaborate? In our earlier case study, we mentioned CodeBuddy, who now organize their operations around AI—task assignment and resource allocation are handled by AI, not managers, as the AI monitors each person’s workload. What’s your current setup like?
Wang Shengjie: When we develop WorkBuddy, we break down all functional modules and let AI define the boundaries of each small feature. For example, I’m currently working on a new module—the “AI Colleague” module—where your team is responsible for closing the loop on all upstream and downstream components involved. All you need to do is collaborate with other teams; for instance, communicating with the WorkBuddy base layer or interacting with upper-layer solutions or the Skill layer, each of which has a corresponding team handling the connections.
This interconnected process involves one AI assigning a contract to another AI: within this module, there must be upstream and downstream agreements, which everyone can review together. It’s not just AI doing the review—human involvement is key during the early review stages. Once the task is received, three small teams are formed to handle different upstream and downstream collaborations. The core functionality is fully cohesive, with approximately three to five developers working alongside PRD and product managers to clarify requirements and finalize documentation.
Now everyone has their own role, and it’s not just one single role—developers can write product manager documentation, and product managers can write code; the boundaries between them are no longer so rigid. The goal is simply to make the output clearer for AI and then let the AI get to work. As the AI works continuously, humans monitor the process and receive the output—most of which is correct, because extensive discussions have already taken place upfront, including collaboration with other upstream teams.
Wang Shengjie: We will create many groups to serve as channels—human-to-human communication will rely heavily on groups. We’ve been thinking: Could there be an AI present permanently within these groups to understand the context? From a technical standpoint, this is still quite challenging due to the high level of noise—AI struggles to clearly determine the timeline and coherence of interactions among members. However, we’re making many attempts, such as compressing and summarizing conversations as supplementary outputs to PRDs, ensuring context is present from the start. We also store these within our code repositories—when AI writes code, it incorporates its accumulated experience; after repeated iterations, it eventually produces a satisfactory result. We then have it conduct a retrospective to update its knowledge, continuously improving the repository. This represents a highly complete end-to-end success story.
Next, you can use this case as a foundation for the next step—such as developing an "AI Team" feature, which is the next evolution of "AI Colleagues." You can leverage the previous experience as a source, allowing the AI to understand, "Ah, so this is how you work." You can also switch to a different model—since models are becoming increasingly expensive, you could use a domestic model or another alternative. Start by having a highly capable model generate a result that satisfies humans, and then the momentum will begin. Anything created can then be reused by other teams.
Yuan Xiaohui: Jason, that was a great share. Since Yi Chuan just mentioned multi-agent collaboration, attention protection, and how teams and multi-agents work together—there are currently no particularly strong productivity tools in this area. How do you address this in WorkBuddy?
Wang Shengjie: It can actually be done better. Once everyone has their own AI, the collaboration among AIs, among humans, and between humans and AIs can all be improved. We’ve made some attempts—for example, I assign one AI primary control to delegate tasks to other AIs, which then complete their tasks and report back. This is essentially a multi-agent workflow.
But the core issue here is: How can we ensure quality in collaboration between AIs? Our current approach is to conduct thorough human reviews upfront, clearly define the agreements, and then have the AIs execute according to those agreements. During the AI execution process, humans perform sampling checks to ensure the output meets quality standards.
Yuan Xiaohui: Alright. Regarding organizational change, we have another important topic—employment. Since AI has led to smaller teams, many people are concerned about their jobs. Ruoyu, what are your thoughts on this issue?
Wang Ruoyu: I think it's helpful to categorize people into three groups.
The first category: those who have successfully transitioned—congratulations, your future is bright. The second category: those who haven’t transitioned yet but are willing to learn and accept change—these individuals are completely fine; it’s just a matter of speed. Some can make the shift in a few months, while others, even at a young age, refuse to embrace new things.
But there’s one group that faces tragedy—if you’re不幸ly part of it, you must switch jobs immediately. What is it? It’s when you realize you’re working hard to drive change, even influencing your colleagues to improve things. Unfortunately, due to systemic issues within the organization, your work is heavily dependent on other departments—your outputs rely heavily on their reviews, inputs, or deliverables, yet those departments have zero interest in change. At this point, I strongly advise you to either transfer to another department or leave the company entirely—don’t stubbornly stick it out. You’ll find that the harder you work, the faster you’ll collapse in this makeshift setup, because everyone will blame you for every failure—you’re the one doing the most work, and therefore, the one making the most mistakes.
So if you不幸 belong to the last category, leave as soon as possible. This group is the most tragic—I’ve seen countless talented individuals lose everything in this situation.
Yuan Xiaohui: There’s a saying that “Choice is more important than effort”—what he just said is essentially encouraging you to make the right choice.
Wang Shengjie: Many members of our team are using AI, but based on my observations, the level of usage varies. As the product lead, I generally provide guidance or accompany them in their growth—helping them learn how to use AI tools effectively.
Here’s a simple example: Some people treat AI as just a web search tool, but AI does more than help you search the web—it can also help you uncover inspiration in reverse. When humans encounter problems, they need to identify them first—but often, the real issues are hidden, hard to spot, or take a long time to find. We can use AI to help you discover these problems or assist in uncovering them, then research them thoroughly and generate insights. Human minds produce ideas at different times, so share those fragmented inspirations with AI, and let it help you summarize or connect them.
Based on your context memory, AI can help you quickly refine your questions and improve your initial needs, moving beyond mere web searches—instead, it more precisely corrects previous misconceptions and shifts error detection further left.
For example, take a simple case: Often, AI gives me long paragraphs of summary. I want to tell it, "You don’t need to interpret it further—just generate a hand-drawn sketch with three options, A, B, and C, and include the key issues I’ve identified." It truly can create the visual feeling you’re looking for, present you with options, and then you simply respond with "This one" or "Not this one"—through image-text dialogue, helping humans quickly understand, identify problems, and ultimately confirm: "Regarding this issue, generate me a final sketch, prototype, PDF, or MD file." My human collaborators and I agree: "Go ahead and start." This is incredibly efficient.
Yuan Xiaohui: Excellent sharing. Currently, many people use AI in a simple Q&A format—often asking a brief question and receiving a lengthy response that may not address what they actually want. After several rounds, attention tends to wane. A better approach is to let AI ask you questions, engaging in multi-turn interactions to better understand your thoughts and intentions.
Actually, I believe everyone should try to understand techniques related to AI use, including prompt engineering, because it determines how you harness this intelligence. This morning, we talked about “intelligence as a service”—for the first time in human history, an individual can access the collective wisdom of all humanity. How you use it creates enormous differences and benefits.
I’ll ask one final question—today’s audience may not yet fully appreciate the capabilities of cutting-edge AI. Do you have any recommendations? Many still feel they need to take a course first—what advice would you give to today’s audience?
Hu Yichuan: I think the most direct approach is to learn while using and use while learning. Since AI itself is an excellent learning tool, when you engage with it deeply, you’re essentially engaging in a collaborative learning process. My advice is: everyone should truly integrate AI deeply into their learning, life, and work.
Wang Ruoyu: My only advice is—whether you want to start a business as a founder or seek a job and join an organization, it all comes down to first principles. If you’re starting a business, make sure you’re the person who can drive the AI transformation, or an AI-native individual, and ensure you have a great vision—I didn’t mention vision earlier because I ran out of time.
If you're joining a company, make sure its leader deeply understands AI and has a clear vision—ensure this. If you're starting your own company, doing this will at least ensure high efficiency and attract top talent; if you're joining a company, it will protect you from being drained by a poor leader or toxic organization.
Yuan Xiaohui: I think you could say a few words about the vision, because the vision is crucial—if the goal of organizational transformation is simply to lay off people or render them obsolete, no one will be willing to drive such a transformation. What exactly is our goal in using AI?
Wang Ruoyu: In this era, vision is more important than ever. In the past, vision existed to unite hundreds, thousands, or even tens of thousands of people around a common goal through a grand narrative. Today, although teams may appear smaller, some question whether vision still matters—I believe it has become even more critical, because products like Laiye Technology or WorkBuddy empower what we call super individuals, as Jichuan and Xiaohui just mentioned.
The value of a vision lies in ensuring you can truly recruit super individuals. Imagine: a super individual—someone who can manage 100 agents working in unison—won’t come to you just to complete tasks and earn a salary. These people are driven by a grand mission to change the world. When you have the ability to attract them, it’s not about how much money you can offer, but rather whether your story, your vision of the world 10, 30, or even 50 years from now, aligns with their beliefs and imagination. The resources you hold give them leverage and allow you to move forward together. This is why vision is far more important today than ever before.
In the past, you could say, “I’ll use capital leverage or various forms of leverage to hire a large team of hardworking individuals who focus only on the present, not the future—then overwhelm competitors through massive productivity and scale.” Today, that’s no longer possible—because manufacturing has become cheap. Your only real advantage is seeing something fundamentally different 30 years ahead. Today, with three people working at a hundredfold productivity, you can directly transform the world.
So, combining what I just said, the two most critical factors—whether you should join a company, follow a boss, or start your own business—are these: first, whether the CEO can drive the initiative forward and truly understands it; second, what their vision is and whether you believe in it. With just these two criteria, everything else is tactical—just minor challenges on the journey to obtain the scriptures: fighting one demon, taking a detour on the road—it doesn’t matter.
Yuan Xiaohui: Regarding vision, I think what Ruoyu just described is a typical male entrepreneur’s grand narrative of the future—spanning 30 or even 50 years. But from a woman’s perspective, we deeply care about the impact we have on those around us and the subtle changes we can bring to society, even if they’re small. Today at lunch, a classmate mentioned that he believes it’s important to teach children how society functions. That small influence—on the people around us and on society—is a meaningful vision for everyone. Just now, we talked about many problems in society; if you can, through your efforts, solve even one or two small ones, I think that’s already a wonderful vision.
Wang Ruoyu: I completely agree. In fact, this is the most difficult part of entrepreneurship. When you first start a company, you can’t possibly imagine such a grand vision—like building a century-old enterprise. That vision evolves over time; it’s a continuous variable. So this is truly the hardest part of entrepreneurship: your starting point is always very small, but the sooner you recognize the larger vision behind it, the more likely you are to attract top talent.
Yuan Xiaohui: Actually, even without starting a business, you can still solve problems—these issues exist in large numbers. Jason?
Wang Shengjie: First, AI is essentially clay in your hands—you can treat it as your partner, helping you solve many problems, and shape it into whatever you desire. You can view it as a small app or a treasure trove for solving any issue. Tell it what you need, let it understand you, and it will transform into gold in your hands. How you shape it depends entirely on you—meaning, the upper limit of AI’s potential is determined by the extent of your imagination. The same applies to our WorkBuddy product—we focus on solving problems guided by imagination and problem-solving.
In everyday scenarios, AI can also greatly help solve many problems, especially in educational settings—we are actively exploring this area. In this context, it’s more about empowerment, learning, and mutual support between people and between people and AI. For example, previously in a library, you might have wished for a teacher or mentor nearby to continuously encourage your learning and understand your progress—knowing what issues arose from your past feedback and how to trace back and reinforce your knowledge. AI serves as an excellent supportive tool in this space, and it’s a field I strongly believe in.
Yuan Xiaohui: Thank you to the three of you for your insightful remarks. Previously, I was very confused and thought the future might be a polarized one—I was anxious, fearing that individuals leveraging AI to its fullest could replace those who are left behind. But once, while speaking with an entrepreneur, he said: “Look, consider the buildings in our world—do you find them all beautiful?” I replied that certainly there are unattractive elements, and our living environments certainly need improvement. In fact, our world still has countless problems that have gone unsolved and unaddressed. And everyone wants to live better—that desire is real. So as long as demand exists, the opportunities for the future are enormous!
This article is from the WeChat public account "Tencent Research Institute" (ID: cyberlawrc), authored by Tencent Research Institute, and is published with permission from 36Kr.