Zeng Ming: The Brutal Truth, Companies That Will Disappear, Why “Excellent” ≠ “Great”

Name: Zeng Ming: The Brutal Truth, Companies That Will Disappear, Why “Excellent” ≠ “Great”
Duration: 154 min 18 s
Channel: Zhang Xiaojun Podcast
Description: Full transcript of Zeng Ming: The Brutal Truth, Companies That Will Disappear, Why “Excellent” ≠ “Great” by Zhang Xiaojun Podcast.

Zhang Xiaojun Podcast· 2 hr 34 min· 25,766 words· 117 min read· English (United Kingdom) ·Watch on YouTube

This is the full transcript of Zeng Ming: The Brutal Truth, Companies That Will Disappear, Why “Excellent” ≠ “Great”, published on YouTube by Zhang Xiaojun Podcast. Every paragraph carries the moment it was spoken, so you can click any line to jump straight to that point in the video, search the whole thing for a word, or copy it out.

0:00（English subtitles translated by AI, for reference only.）Hello everyone, I'm Xiaojun Today's guest is corporate strategy expert Professor Zeng Ming Professor Zeng is the former Chief Strategy Officer of Alibaba Group He also helped found two business schools Namely, Cheung Kong Graduate School of Business (CKGSB) and Hupan University But the identity he is better known for is as someone who has thought and researched deeply about corporate strategy Many founders and entrepreneurs say they have been inspired by him As a researcher who has lived through the entire internet cycle We talked with him about his complete thinking on the new era and new paradigm of AI And he drew comparisons with key periods in history

0:37Listening to him gave me a more holistic perspective So next is my interview with Professor Zeng Looking forward to 2026, when we progress together with AI The corporate system of hierarchical management will decline and die out But it may no longer be called a company It definitely won't be called a company It definitely won't be called a company So companies will disappear Companies will die out There is nothing to regret about that There will be more fun things in the future Tell me, how many people are happy going to work at a company these days? They're not happy Let me put it a bit sharply: For this generation of entrepreneurs

1:15there is a big hurdle to overcome Because the basic law of economics is as long as there is homogeneous supply you cannot earn high profits You are just a company that isn't worth much Google won't necessarily be the Google of the next era Will ByteDance be the ByteDance of the next era? Likewise, most likely not ByteDance's models are doing quite well today At least they're in the first tier in China No—so ByteDance has a big chance of becoming an AI cloud company But that big to-C application carries enormous uncertainty You mean Doubao? Doubao definitely isn't it Otherwise they would be pouring money in regardless of cost

1:57They stopped because Doubao isn't good enough Not enough of a future Will social relationships be restructured? Is Tencent safe? They will definitely be restructured Tencent, supposedly the safest— why has it been so panicked lately? Precisely because it might not be safe either Relationships between people will be restructured People's social relationships will be restructured And what about the relationship between people and goods? That will be restructured too I think Alibaba faces the same challenge In the face of this giant wave which giants are safe? I've always believed that before a giant wave no one is safe Do you think that by going all in on AI they can still become companies of the next era?

2:31I mean, era-defining companies? The challenge is enormous Looking at history is there any precedent? Companies from one era transitioning into the next era and still doing exceptionally well— are there many? It's not a question of how many It's a question of whether there are any at all Any at all? Basically none So this doesn't depend on your strategic choices No matter what, it's hard to stay at the table That's the despairing part That's not despairing at all People are born, age, get sick, and die Companies also are born, age, decline, and die Nothing is more natural than that That's why new companies get their chance That's why young people can have dreams

3:08Professor Zeng, I've heard many entrepreneurs mention you They say a very important strategic transformation they made They say a very important strategic transformation they made may have come from taking your class at Hupan University There are many business school professors I wonder why so many entrepreneurs say they've been influenced by you What do you think makes you different from other business school professors? First of all, I think I was extremely lucky I had the opportunity to join Alibaba when the internet was growing fastest And I took part in a phase of hypergrowth for a company So that experience combined with my solid academic training allowed me to integrate theory with practice

3:46Because while I was doing the work, I often thought about which of the things I used to teach could actually be applied And in the process I would also distill what new things could be turned into a theoretical framework that could still be used in the future So combining theory and practice was a major breakthrough And second, I have continuously studied frontier industry development So when I talk with entrepreneurs besides discussing strategic methodology with them I can also explore with them how the industry will actually develop next and what kind of strategic decisions might be more appropriate So combining these two is a fairly rare capability

4:26Take Alibaba as an example Because this is the company you were personally involved with the longest Because this is the company you were personally involved with the longest Because this is the company you were personally involved with the longest Over the past 20 years, were there strategic decisions that truly determined Alibaba's survival? Which ones would they be? I happen to be doing a retrospective lately So I'm quite familiar with Alibaba's first 15 years Because I participated in it fully In fact, the most important decisions were probably three Founding Taobao Founding Alipay Founding Alibaba Cloud And these three decisions were not only unrecognized by society at the time

5:00The controversy inside the company was also extremely intense People may already know a lot about the controversy over Alibaba Cloud But when Taobao announced its founding several executives were very strongly opposed Why? They felt Alibaba's original business hadn't matured yet And then you went to attack eBay's core main business And then you went to attack eBay's core main business The risk was too great So these are the three most important decisions Then there were the slightly less important ones Like how Taobao went from a platform for buying and selling goods to becoming an ecosystem supporting the development of e-commerce And then how to go all in on mobile

5:42I think these two decisions were actually very important too But perhaps not as significant But as I just said, I should add one more point After you make a strategy there are still layer upon layer of decisions Secondary strategic decisions are also very important For example, in Alibaba Cloud's early days do you serve only one scenario or should the architecture support multiple service scenarios from the start? Should the scenario you serve be Taobao or search? And open source or closed source? These were all second-level strategic questions with huge debate And further down, your database changed Should you build your own database? Then computing also changed And storage changed too

6:28That's how the slogan "de-IOE" came about back then That is, replacing IBM, Oracle, and EMC (storage) So just like today, once computing is solved you find storage becomes the new bottleneck So in the process from strategy to execution there are many very important decisions The further down you go, the more strategy and tactics are tightly intertwined So making one big strategic decision isn't enough From decision to execution to finally seeing results there is an enormous challenge At the moment of making a strategic decision how did you know it was a good strategy? Because there was so much controversy at the time Generally, because you make many strategies

7:11how do you judge that it is a good strategy? We didn't know Because we believed, we saw it All the truly great companies that became era-defining were eventually proven right in their contrarian view Mm Of course, in the process there were also many crazy ideas that simply disappeared This point must be emphasized It's not crazy ideas What we see is the so-called survival bias You are seeing the bias of the winners The most successful companies are necessarily contrarian But being contrarian doesn't guarantee success But at the time based on our own capabilities what we could see and were willing to believe in was that picture So strategy, to a large extent

7:54is actually about how to create the future It's not simply a plan Can only contrarian strategies achieve great success? Can a consensus strategy achieve great success? I'm in a consensus track But I do it better than everyone else Can that achieve great success? That's excellence That's excellence Not greatness Yes Greatness must come from being contrarian Yes Now, in today's AI era we have entered a brand-new cyclical paradigm phase As you said, this is a new civilization Do you feel lost or anxious? As someone like you who formulates strategy when you look at the future how do you feel at a moment like this? It is indeed very confusing

8:32Because you know many new things will certainly appear Because you know many new things will certainly appear But you don't know which are fleeting little ripples and which are the big waves representing the future What will change and what won't This is extremely confusing Which do you now see as small ripples and which as big waves? Do you have some preliminary thoughts and judgments? That's why I wrote this book ("Intelligence: The Essence of Business, Organization, and Strategy in the AI Era") Because I'd been pondering this for the past few years I often had observations and insights that seemed pretty good But they were quickly overturned But in the first half of this year it became clearer

9:08Some trend signals began to converge Similar signals began to appear And then I saw a possible framework that could explain many things happening now and even project the next steps of future development So at that point I became less anxious Because I roughly understood which development stage we're in And knew how things should be done Right—the reason for our conversation today is also that you recently published a new book, "Intelligence" You also said there was no writing plan for this book at first It emerged naturally Through a series of reflections over the past half year you figured out many questions Could you introduce this book to everyone

9:55as well as your creative process? My own profession is strategy So what I discuss most with entrepreneurs is facing such great uncertainty in the AI era how do you do strategy in such a rapidly changing environment? So this was something I gradually figured out But what really made the whole framework emerge was that I felt I roughly understood which development stage AI industrialization has entered and the different operating rules of each stage I think once this is clear for roughly the next three to five years it provides fairly clear guidance So what stage has AI industrialization reached today? How do you divide these complete stages? I actually did a fairly systematic study

10:40From mobile internet to PC internet And then to the PC And when I was working on cloud computing I actually studied the First Industrial Revolution, the Second Industrial Revolution and the Third Industrial Revolution—the entire history So broadly speaking, the emergence of new technologies I'm talking about disruptive innovations of massive general-purpose technologies actually all went through three stages Very interesting The first stage is the technology gradually being accepted by society And then it becomes the infrastructure of society That's the first step The second step: based on this infrastructure a massive number of new applications emerge It's an era where a hundred flowers bloom The third is when these massive applications have appeared

11:29people realize we need more fundamental methods to run this new world This enters the third stage I call the third stage the native application stage Take mobile internet as an example By the time of Douyin and ByteDance, it was the native application stage Yes In the early days, for example With mobile internet, young people today may rarely have even heard of the BlackBerry But the BlackBerry was the first to be called a smartphone And it held a monopoly position for a long time And its market penetration was very high Even as late as 2010 it still had over 40% of the US smartphone market While the iPhone had only twenty-some percent

12:18It was actually only after the iPhone 4 that the iPhone started to pull ahead and take off So actually this first stage, second stage And after the iPhone came the App Store After the App Store, a new generation of native applications appeared And in the early days, actually Back then I was investing in the internet, in mobile internet There were apps like Moji Weather and such A bunch of these apps quickly surpassed 100 million daily active users But you'd quickly discover that they didn't represent the future The mainstream applications that solve core needs are the true native applications Why must there first be an explosion of applications

12:54before native applications arrive? Why aren't native applications born in the second stage? Because the explosion of applications gives direction to the technology's next stage of development Once technology becomes infrastructure, it doesn't stand still It needs to find its new direction of development But which direction to develop in is actually unclear Just like ByteDance, as we just mentioned Just like ByteDance, as we just mentioned they actually built Toutiao first And even before that, Neihan Duanzi a bunch of small products Because in the early days of that technology you don't know what will work So you have to try different things In the first stage Only text was mature at that time

13:32Which is why Toutiao appeared But Toutiao pioneered the recommendation algorithm a new thing that was both a technology model and a business model Then, with bandwidth computing power, and development in all these aspects short video emerged Then this new format of short video and the underlying technology of recommendation algorithms once these two were combined a phenomenal native application like Douyin emerged A truly national-level application appeared So between native applications and infrastructure maturity there's a transitional stage in between Can we call the second stage the transitional stage? If you put it that way a lot of entrepreneurs in the middle would be heartbroken That's why I hesitate

14:19Because the second stage itself is very exciting and very valuable That's why I say the native stage is the third stage I actually want to tell everyone that there may still be eight years, or ten years of very exciting time belonging to the second stage You can't leapfrog into the third stage Without the accumulation of the second stage, you can't cross over You just can't get to the third stage And based on the accumulation of the first stage what should be done now is what the second stage calls for the great explosion of applications In this AI industry cycle we're in now have we finished the first stage

14:55or are we still in the first stage, or at the start of the second stage? Exactly at which point in time are we? I think 2026 should mark the first stage the symbolic point where it is basically complete The core reason is that at the beginning of this year you may remember the word "token" a broad consensus formed There were two or three months of very intense discussion Jensen Huang said at NVIDIA's GTC conference that NVIDIA's future is the token factory And Sam immediately said OpenAI's core business is selling tokens So why is the token so important? The reason this new term "token" (词元) got defined

15:35is that once a technology has a standard unit of measurement it can be used at scale, in a standardized way So today we talk about a kilowatt-hour of electricity a ton of water, a GB of data These let us very simply use a technology that was once at the frontier Today everyone knows: how many tokens I use how much work I can get done how much it will cost me At that point it becomes society's infrastructure So this is the tail end of the first stage a sign that the first stage is maturing It will continue to develop The services will get better and better The intelligence will grow stronger and stronger

16:14The cost will get lower and lower But it has formed a matching business model And become a new concept everyone can accept So this stage is mature Have we reached the second stage? The interesting thing about this year is that it's also the opening of the second stage Because around this Spring Festival, something else happened that got everyone especially excited OpenClaw The lobster—everyone was "raising lobsters" People didn't sleep well through the Spring Festival Frantically tinkering with these things every day What did the lobster show everyone? What they actually saw was an Agent capable of independently completing tasks— the limitless potential of Agents That's why everyone was so fanatical

17:00So I believe the second stage AI is entering is the agent stage Agents are the new applications of the AI era So the first stage has matured And we now stand at the beginning of the second stage Yes If we draw an analogy say, to which year of the internet or which year of the mobile internet can we make such an analogy today? Of course this is a bit like carving a mark on a moving boat but we'd still like to make the analogy Right Yes Well This is a question I've been pondering for two or three years Actually, two years ago I gave a talk

17:32called "Let's embrace Yahoo" Yahoo is coming soon And then at the beginning of this year I felt a bit torn myself How did we retreat back to the browser stage? It seems even earlier than Yahoo So your judgment moved even further back Yes, it moved further back Because when I saw OpenClaw and these things come out my first reaction was we're still waiting for a browser to come out This isn't even after the browser came out We're still waiting for a browser to come out Why did you have such an insight? It's because what appeared alongside the Agent is the skill The capability So when we think about the essence of an Agent

18:10it's actually about sharing capabilities Right? We package the capabilities we know and share them so others can use them too And OpenClaw provided a framework and an approach So capabilities can be packaged and invoked this way It suddenly occurred to me what was I most fanatical about in '92? Building a website Right? Back then it was all about building websites Everyone wanted to build a website And what was the purpose of building a website? To share information So this round of AGI innovation takes us from the information age into the capability age The massive amounts of information and knowledge have all been processed So sharing information is no longer necessary today

18:56Everyone just asks ChatGPT for answers directly People no longer search for information But today's large models achieved this because of the website-building movement of '92 which got everyone to put information on the internet completing the digitization of information So the key to making analogies is knowing when an analogy isn't rigidly mechanical It's about understanding, after abstraction and modeling, the underlying laws of how it operates rather than getting fixated on the superficial form of the technology So when I thought about how everyone was building — one building websites, one building Agents — and one is about sharing information while the other is about sharing capabilities I felt that structurally, it all lined up

19:35Why was the browser important? In that era, it marked a new stage The browser fulfilled two core functions Through HTML and all that, it established, for building websites, a very simple, unified standard So building a website no longer had any barrier Anyone with information could build a site Before, building a website was specialized work for professional engineers So it dramatically lowered the barrier to creation That is, what we call today the barrier for developers was greatly lowered And on the other hand, it reduced the barrier to browsing websites to zero Because you only needed to click to complete the whole process of browsing information So the user side suddenly exploded

20:19And then the supply side exploded too And then endless websites kept emerging every day And then people didn't know what to look at anymore That's when Yahoo appeared In the beginning, Yahoo was just a directory telling you among the 100 websites that appeared today, this one is the most interesting Later, when there were 10,000 websites, it would tell you what the top 10 most interesting ones were And then it categorized them for you Entertainment, news, stocks — it started classifying So Yahoo was actually a process of structuring information Right, it was just a directory It was reducing entropy, not increasing it, right? While entropy increases, there must also be entropy reduction

20:55To lower everyone's cognitive load And today we're still in a process of entropy increase That is, an explosion of all kinds of skills and capabilities Right, because if you casually ask anyone around you, they can't readily name two or three Agents they use every day So who do you think has the potential to do this entropy reduction? Have you seen any embryonic form of Yahoo? Is it something like OpenClaw? No, I haven't seen any embryonic form of Yahoo at all That's why I said even the browser hasn't appeared yet So if you ask me to make a wild guess, I think the next biggest opportunity is an opportunity like the browser

21:30It will unify the standards for Agents And at the same time let users, with very simple methods, call upon Agents And then there will be a massive explosion of Agents This might be, in the next two years, possibly the most exciting thing Is there really no such embryonic form today? So this is also a contrarian view, right? Everyone seems to think the "lobster" thing was also a flash in the pan It seems nobody talks about applications anymore But what I'm saying may be very optimistic But I am indeed saying that we extrapolate based on history to form an abstract model for thinking about the future And at the underlying level, what I can see

22:12is exactly such an evolutionary path So in that sense, I'm no longer anxious I just wait for these things to happen So imagine what such a product form might look like What might its prerequisites be? The interaction form can't be imagined That's up to everyone to create But we can summarize an abstract law Yet we definitely can't deduce a concrete form It was still a few years from Yahoo to Google Since you haven't seen Yahoo, you probably haven't seen this era's Google either, right? That's a question I can't answer right now Because I'm not quite sure whether our era will necessarily have the huge gap between Yahoo and Google

22:50Because Yahoo ceased to exist later on This, I think, I can't see clearly now I also think it's too early No need to discuss it Let's wait until we see Yahoo, let Yahoo shine for three years first, and then we'll probably be able to see who wants to challenge Yahoo Yahoo was founded in '95 and was glorious all the way until 2000 After Google was founded, it was a To B service provider for Yahoo It provided them with an algorithm service Just a little brother under Yahoo helping it with technical support, that's all In this sense, Yahoo gave birth to Google The key variable of this era is the model

23:29How do you understand models? What is the essence of a model? What can it be compared to? There are many possible analogies For example, people say data is the oil of the new era The model is actually a refinery From the angle of broadly providing infrastructure, it's also a bit like another company that has disappeared, AOL It was an internet access service provider It enabled everyone to get connected to the internet That in itself was a huge undertaking And today, large models let everyone easily access an entry point to intelligence So in this sense, it's also an infrastructure service provider So you think model companies are infrastructure service providers

24:15What companies can they be compared to? For example, today's OpenAI, Anthropic, Kimi, DeepSeek — these companies To put it bluntly, I think they're AI cloud companies They are the AI cloud companies of the future That's why right now, for example, the game between these large model companies and the cloud vendors has become very interesting and complex in the US Because the cloud vendors feel they've given too much and gotten too little And the model companies very likely believe that in the future they'll build their own cloud services or even develop their own chips Companies like OpenAI So between these two ecosystems, there may be some collisions and reintegration

25:01And then a few relatively stable AI cloud companies will form That much is certain As for whether large model companies have a chance to go from basic service providers to successful application companies, I think that's an open question If I were to give a subjective conclusion, the probability wouldn't be very high Why? There's no historical precedent Codex is doing very well now Claude Code is doing very well No — that's a foundation of large models, an extension Coding has to become part of foundational intelligence Without coding, we couldn't even talk about calling APIs today You could say tokens wouldn't make sense It's because of this coding capability

25:44that tokens became so valuable Because its level of intelligence — its reasoning intelligence — made a breakthrough But look at a company like Anthropic — it's been around for less than 10 years In just a few years it's already valued (slip of the tongue) at a trillion dollars That's extremely rare in human history How do you explain this phenomenon? That description itself is an illusion and a misunderstanding Because young people tend to believe that this time is different in every era That's the classic "this time is different" Actually, I specifically looked into this Going back to what we just said about Yahoo Yahoo was founded in '95

26:19By 2000, it had reached a market cap of $120 billion In five years What does that amount to? A $100 billion market cap back then is roughly equivalent to a trillion dollars today If you factor in inflation and the Nasdaq's baseline growth, that $100 billion back then, at roughly an 8% annual return, equals a trillion-dollar company today So reaching $100 billion by 2000, in five years — the shock it gave people was no less than Anthropic today It was equally shocking And the vanished company I just mentioned, AOL, started providing services in 1988 At its peak in 2000, it was valued at $220 billion Higher than Yahoo

27:08It was the most valuable company at the time And then it went and acquired traditional media, acquired Warner, reaching over $300 billion So that's also a staggering figure And Google was founded in '98 and reached $100 billion in 2005 BlackBerry — at its peak, it was also a hardware manufacturer — was the fastest company to reach a $100 billion valuation So actually, when a technology has such a massive impact, the first wave of companies that break out successfully carry this much value So what I mean is, Anthropic and OpenAI — this is remarkable, a technological breakthrough and a commercial achievement But precedents like this in history

27:53have happened at similar stages But they may not survive to the end They'll likely keep doing quite well for a while, but most probably they won't be the big players, the big winners of the native application stage Even going back to the oil analogy we talked about, refineries and chemical plants are actually two different types of businesses Chemical companies later became far more valuable than pure refining companies Chemicals are the applications Refining is the most basic thing — extracting the oil and turning it into base fuel And there's another analogy — take cars, for example A car definitely can't run without oil But an oil company definitely can't make cars

28:38You haven't heard of any car company that evolved from an oil company I happened to read yesterday something Dr. Lu Qi said back in 2024 — OpenAI, as an AGI company, will definitely be bigger than Google in the future It's just a question of whether it's 5 times or 10 times bigger Based on what you just argued, I feel you not only don't think OpenAI could be the Google of the new AI era, you even think they might just be a passing company in history They might even disappear — is that your view? That is my view I believe there will definitely be a $10 trillion company in the future

29:13But it may not be the two companies currently in the lead Actually, what I understand Dr. Lu Qi meant is, in the future there will be a Google of the AI era and it's very likely to be 10 times bigger than today's Google No, he was talking about OpenAI I understand But back then, people could roughly only imagine that OpenAI would represent the future But purely because I'm a researcher of business history — lately I sometimes joke that because in Silicon Valley, the most fashionable word now, or the highest-status word, is "researcher" The moment you say "researcher," everyone respects you And I think I can now also claim

29:50that I'm a researcher too I just study the business world's laws of dynamics And what we can see — the theoretical framework I can see tells me that companies of the first stage can hardly survive into the second stage And companies of the second stage can hardly survive into the third Why is it hard for first-stage companies to survive into the second stage, and second-stage companies (hard) to survive into the third? Logically, they entered this market so early, and they're the most sensitive to cutting-edge information Why can't they transform in time? For example, just switch right away from being an infrastructure company to becoming an Agent company

30:23That seems smooth enough I think it's not smooth Because first-stage technology — its core is whether it's usable at all So it involves a lot of original technological innovation The marker of the second stage beginning is that first-stage technology development gradually slows down And then it becomes a technology everyone can call upon I no longer need to reinvent the wheel And second-stage applications actually require strong insight What they demand is products with strong originality and that kind of empathy for ordinary users And then you need a bunch of new things in application technology So you could even call that bunch of new application technologies the large model of the new era

31:08It actually involves new innovation at many technological frontiers For example, take context management — that thing There's a lot of technical innovation to be done there Whether you define this as a large model or not a large model isn't actually that clear-cut And beyond that, actually, it also includes — AI will also see many new paradigms emerge For example, many new neo labs are appearing now Inside these new neo labs, some of the people may turn to supporting the new technologies that applications at this stage need The combination of these two may become the king of the second stage So technology is still advancing But technology may not be growing

31:51along its original trajectory in a linear fashion There may also be a paradigm shift in technology In your view, whether in China or the US, what kind of state will the landscape of model companies ultimately form? Very simple For a public infrastructure industry, only one business model holds: oligopoly plus strong government regulation Because the whole society depends on it, the government will definitely regulate it heavily Second, because this supply is so important, there can't be just one supplier And because this infrastructure requires such huge investment, there definitely won't be disorderly competition There won't be many suppliers So it's classic oligopolistic competition, with strong government oversight ensuring society has stable,

32:36low-cost supply That's what infrastructure means, right? Can water and electricity prices be raised at will now? Of course not I know this reality is harsh for many people starting businesses now For example, OpenAI's, Anthropic's, Google's models — do you think they'll all coexist? Is there a hierarchy of better and worse among them? Of course there are better and worse ones This is the most valuable competition at this stage The industry's endgame I talked about doesn't negate the importance at this stage or what makes these companies remarkable And this year especially, with Anthropic taking the lead first, OpenAI catching up, and Google being left behind for a phase —

33:13it's actually that organizational capability and technical capability have opened up some gaps And players like xAI and Meta have simply dropped out of the race In the process toward oligopoly, some have already exited, fallen behind Right, fallen behind That's also what I wanted to ask you Why did xAI and Meta fall behind? They seem to have plenty of resources Their bosses also attach great importance to it You could call it a top-leader project This may relate to what I discuss in my new book, "intelligence compounding" — that basic concept I think Anthropic, starting from the second half of last year, maybe through the first half of this year,

33:49made people clearly realize that they've started using large models to train large models In other words, AI has entered their core business — large model training They applied the tool of AI to their own core business So their model training capability improved dramatically That's why, in the first half of this year, everyone got so excited Because they discovered that large models can still improve performance this much at such a fast pace This is because the large model itself has entered its own training process But it hasn't yet reached — there's that buzzword again — the fully self-driven self-learning It hasn't reached that stage yet It's only been used partially

34:35But it has already generated great value Actually, these two companies also have organizational advantages Like what has been discussed just as much this year — there are many articles discussing Anthropic's organizational advantages and strategic focus And these are areas where, relatively speaking, the other companies may be weaker So when your organizational advantage combines with a new way of operating with technology, it produces a huge leading edge And this leading edge has really already started throwing the weaker players off the back What about the Chinese model companies? How will they evolve further? China and the US maintain two somewhat different ecosystems I think Chinese model companies will keep

35:19catching up and developing But competition among Chinese model companies will also be somewhat similar to the US companies Among these companies, big and small, in the end — say by the end of this year — whoever fails to reach a 3T-parameter model will probably be out It's an elimination tournament Just like the World Cup now You could say it's the round of 8 into 4 right now And then 4 into 2 At the 8-to-4 stage, elimination will very likely begin So in the end, only a limited few will remain So in the end, only a limited few will remain and become an infrastructure That's your view

35:54Yes As it looks now, who has been eliminated? We know about the US What about China? For China, it really comes down to year-end The year-end 3T — right now everyone is going all out on exactly that K3 just reached 2.8T Which basically means Zhipu is taking the lead And Kimi submitted its paper first at 3T The results are basically satisfactory Next up, let's watch MiniMax Qwen will definitely release something soon MiniMax and the others They're pretty much all running this most fundamental new competitive race Will the model landscape stabilize this year? Not necessarily, actually That is, maybe once the top two or top three have widened

36:36their lead over the rest those two or three players may find it hard to pull further apart That's the black hole theory you mentioned Right To form a huge black hole effect and make it very hard for others to catch up Right. Actually Anthropic and OpenAI in the US have actually started from this flywheel of intelligence compounding Because they will keep moving down the path of self-improving So the gap keeps widening So they've actually formed something of a black hole effect That's also why the American cloud providers are getting a bit nervous Because it depends on what kind of partnership between the two becomes very important in the future

37:15So in your view, has the window for building models closed? Should new companies still build models? For large model companies that serve as infrastructure, as we define it, the window is definitely closed Because competition has entered a stage close to where the intelligence flywheels are already spinning If you come in now, you have no accumulation at all Will these model companies converge or diverge in the end? The characteristic of oligopolistic competition is a certain degree of differentiation but sufficient homogeneity Homogeneity ensures substitutability Differentiation ensures everyone can still make money So is the model business a good business? It's a very good mature business Why do we say companies are companies of their era?

37:59So you definitely won't be the hottest star of the moment In three to five years, if you really haven't entered the next stage of competition, you become a service everyone just calls on People will only complain when it's unstable They won't praise you every day OK You think in the next three to five years we might see signs of this new thing We should be able to see it So next, if they are infrastructure providers, then in the next stage, what will we see? What kind of companies will become the companies of the new era? It depends on who becomes Netscape and who becomes Yahoo Who establishes the standard for Agents

38:37Who captures the user entry point for Agents This view is actually quite non-consensus compared to what we imagine today So if you go to the WAIC conference now, only on the basement level, in a small area, can you see the companies building applications The big players on the second and third floors are all doing things that incumbent companies already do This is interesting. This is a snapshot of 2026 A snapshot of the WAIC conference What do you think it will look like in three years? Take a bold guess I think within three years, an entry point for Agents may emerge Because there are too many Agents

39:10Think about it—let's take an ordinary person's perspective How do you know which Agent is usable? Would you dare entrust something important to an Agent? Would you dare hand your wallet to an Agent? Would you dare hand your personal privacy to an Agent? And where does this trust come from? And how do I know the fees it charges me are real? All these things actually require a possibly a new service provider to build this kind of trust for everyone So that we're willing to say, OK, I trust the standards you set I trust your recommendations And then I trust the Agents on it I dare to use them

39:51Whether they're good or not is my own choice later This is the topic we're going to talk about next which is Agents You think Agents should have already exploded by now, right? How do you see its current state? Right now it's a hundred flowers blooming with everyone showing off their skills Everyone is looking for that entry point Because actually, last year people were already calling it the first year of Agents And there was a lot of Agent entrepreneurship Actually last summer in Silicon Valley, I already experienced the first wave of Agent fever At any demo day, there were basically a dozen or so projects all kinds of Agents

40:29Occasionally a few doing infra building infrastructure But most were building applications But those applications actually aren't doing so well this year That's because last year's applications were really all doing very simple tasks Out of 10 demos I saw in one session, about 3 to 5 were doing advertising Some of them were just generating creative materials Some were just doing ad placement Such simple things At the time I felt they were unlikely to have long-term value But back then, for example, I saw a company starting to work on growth I thought that might have a future Why? Because growth is complex So it has to do complex tasks

41:18Only by doing complex tasks can you build your own intelligence closed loop And then you might build up a bunch of your own valuable technology accumulation rather than building scaffolding That is, when intelligence capability is insufficient, building applications is largely building scaffolding And then waiting for the models to naturally flood over your scaffolding But today, the entry point for building applications and the approach need to be completely different You have to assume intelligence is fully sufficient And then ask what work it can actually do So today's core is getting work done And doing a complex job The simpler the job, the easier it is to be swallowed by complex jobs

41:59Because it has little room to grow Once the job is done, it's done Doesn't it just stop there? Like Manus, which last year said it was building a general agent product form What do you think? I haven't been in touch with them for a while I don't know what their next step is But Manus at the time indeed had a very sharp market sense Because they actually seized the rare window of opportunity from the coding spillover But their positioning is the general Agent So the general Agent what it could reach at the time was the productivity needs of a group of prosumers That entry point was right at the time

42:48Today, for an Agent to stand firm, it needs at least a 10x entry point Only then do you have room to grow What is a 10x entry point? It means being 10 times bigger than that application scenario For example, everyone making PPTs or producing reports—those aren't big enough And today the technology has pushed another step forward The level of intelligence has advanced another step Things like long-horizon planning have all improved a lot compared to before So can you take a big step forward and solve a complex task? Does that require entering a vertical scenario? Should you build a general Agent or a vertical Agent? This is where everyone next

43:31competes on creativity and talent For example, education and healthcare are typical big verticals You just need to cultivate within them Within that vertical, can you find a good enough entry point that you can cut into now and that can grow big in the future? But within the general space, there's also a split between to B and to C In to B, which one is key? That's different again I think by the end of this year, you'll hear a bunch of Agents aiming to be the organizational brain The organizational brain is an Agent The organization's operating system Just different names Operating system Or the CEO's Agent team

44:12This will definitely be this year's hottest main track That's the to B side The to B side What about to C? On the to C side, people are starting to do emotional companionship again And problem-solving ones are all starting to appear Many people say this wave of applications can hardly form the kind the previous internet and mobile internet era had— the data flywheel What do you think of this phenomenon? Do you think they should pursue a data flywheel? The data flywheel has been upgraded to an intelligence flywheel That is, the flywheel of intelligence This is what we call intelligence compounding Today people are also talking about the data flywheel

44:52The data flywheel has always existed Actually for the internet, what's more special is the network effect Let's first talk about the data flywheel upgrading to the intelligence flywheel It's because that is to say, to really get the data flywheel spinning, just like today's embodied robots, you actually correspondingly also need algorithmic innovation and also application scenarios So this is also a three-in-one thing And the intelligence compounding we just talked about also requires you to form a new learning scenario You call upon foundational intelligence but you enable the Agent to possess a higher intelligence to complete a specific task And in the course of that specific task, there will definitely be a data flywheel

45:37Because the premise of the data flywheel is real-world feedback And the premise of real-world feedback is that this Agent works independently without human involvement And in the middle there will definitely be algorithmic innovation Including what we just talked about, in a broad sense, including context management, including memory, including new interaction methods And this forms a compounding of intelligence In this scenario, it becomes smarter and smarter And this intelligence compounding can penetrate and extend into other scenarios to solve more complex tasks That's its development path And whoever finds, or luckily stumbles upon, the most scalable and most generalizable scenario is most likely to form a black hole effect

46:26Then it has a chance to break out and enter the third stage of competition This requires many conditions to be met for a native application to emerge, right? Right Have you seen such good scenarios now? Probably not yet No If there were, it would definitely be a hit Everyone would definitely rush to use it Look, with native, even a phased native product, say the best thing of the second stage, must spread by word of mouth Because it must have a big breakthrough in experience Just like Lobster also blew up on its own No one did any paid promotion And Yahoo was the same Word of mouth made it an instant hit

47:04In '93 I was studying at UIUC, the birthplace of the browser So I was probably among the first ten thousand people in the world to use a browser The whole campus went crazy, staying up every night just using the browser to look at all kinds of things you could see And soon everyone felt, why is there so little to see online Then someone realized making a website and uploading it must be the most interesting thing And soon someone discovered this was actually a business opportunity I can make money And this flywheel started spinning So at this OpenClaw moment, did you feel the same as during the browser explosion

47:39of '92 and '93? People can't get OpenClaw installed People can't get OpenClaw installed Everyone feels very anxious, feeling left behind by the times Yes, yes, yes What we just talked about was the data flywheel What about network effects? What will network effects look like in the AI era? First, network effects will continue to exist Just like infrastructure, they've become a standard feature of this era Because in the end, broad connectivity is still needed Even from the nature of data, the more diverse the data, the more differentiated the data, the more it can generate new insights With the same type of data, once scale reaches a certain point,

48:16its value actually has diminishing marginal returns And second, I think the upgraded version of network effects is actually the black hole effect The black hole effect means the more application scenarios, the more complex your connections, the more likely you are to expand your intelligence So I think the black hole effect, in a sense, is the upgraded version of network effects in the AI era Actually in my book Smart Business, I talked about the two components of the DNA double helix One is data intelligence One is network coordination Actually the black hole effect means both have been upgraded That is, data intelligence— actually today's AI is still data intelligence

48:56It hasn't entered the so-called symbolic stage of AI intelligence We're still in the data intelligence stage And network coordination has actually become a kind of internal network structure of the black hole How many scenarios it can quickly swallow, rather than our original simple connections Will platforms still appear in the new era? Definitely The entry point is the platform Because I think many platforms of the previous generation existed because our information efficiency wasn't high enough So they needed a platform to further circulate information efficiency Today, we can use judgment to replace information asymmetry But judgment is also a scarce resource So as we just discussed, a two-sided market will also emerge

49:45Between massive numbers of Agent developers and massive numbers of users, how is this matching accomplished? Looking at a website, it's easy to judge whether it's good or bad The cost is low If it's bad, you just stop looking But when you use an Agent, there can be very big negative consequences So our ability to judge an Agent's capability has a higher bar So at this stage, we will most likely cede this judgment to an independent third party that we consider trustworthy And this is actually a typical two-sided market structure A two-sided market has a moderate degree of network effects So between the new platform and the old platform,

50:29what might be different? Can old platforms become new platforms? Can they evolve over? Most likely not Because old platforms do simple information matching New platforms need to match capabilities with needs So the complexity is much higher The things on both sides have changed Yes, it's the matching of new species It's still matching, but what's being matched is different So new companies will most likely be born Yes Because if you're not Agent-native, how can you judge Agents? How can you provide Agents with the best, a universal development platform, right? But aren't we still far from the platform stage? It may take a few more years Roughly 3 to 5 years

51:12This year people are also talking about Agentic OS First, what is Agentic OS? Do you have a definition? I don't think anyone can define Agentic OS I place it in the third stage So we can discuss it on that basis Because once you're an OS, you start scheduling resources It no longer does matching You just tell it an intention, and it directly calls on all the Agents in the market to get the job done for you You don't even see the Agent First, Agents have to truly explode Yes So it must go through the second-stage period Right, right, right So only at a sufficiently complex stage

51:49do you need to directly schedule Agents Agents also have to mature to that stage to be simply called upon So Agentic OS may be further away than we think I personally think it's a third-stage thing It's relatively far off But everyone, because in the second stage, it's easy to imagine what the third stage might be So this discussion is meaningful But it most likely can't land at this stage So the Agentic OS you envision, what might it look like? And who might contend for its dominance? It could be model companies It could be application companies, internet companies, compute companies, or software-hardware integrated companies like Apple According to my theoretical framework,

52:24it must be a company that no one knows what it looks like That's what counts as big innovation Otherwise it doesn't deserve its future value So, because, that is to say, what the Agent will look like None of us know right now Right? And then when you get to the second stage what the native hardware for Agents looks like we don't know either So both of these things need to be explored to a certain stage before we will probably find the most natural way to interact with AI And then that AI can really provide us with convenience we can't imagine today That would be the science fiction of the future

53:06We should probably ask science fiction writers what they think things will look like at that point It's interesting that we can roughly based on history project three steps ahead but at each step what exactly the Agent looks like what the Agent ecosystem looks like and what the Agentic OS looks like at the next step what native applications look like are all question marks Yes, right But the benefit of discussing the third stage is you roughly know which direction the final endgame is heading So when you work on the second stage you'll be more targeted And then in the mid-to-late second stage part of people's attention will start moving toward the third stage

53:42But, well this year Agent application companies are all struggling badly You could say we've reached the coldest point for Agent applications in the past two years Last year people were quite enthusiastic about investing in application companies But this year people feel isn't that just a skill of the model? It's become very hard for application companies to raise money Do you think this phenomenon will be short-lived or long-term? Is there anything you'd like to say to the CEOs of application companies? I think right now, in the application space entrepreneurs are in a contrarian phase because there isn't a good application to show everyone how much potential it has

54:18and how it should be done So everyone is groping around Everyone is groping around Take video generation, for example This is already a relatively mature field and there are many companies building Agents And everyone's ultimate goal is probably to build the TikTok of the AI era like Douyin and such But everyone knows that's far from where we are now So what should your first step actually be? This really tests everyone's skill Then AI short dramas and animation have a rather low barrier to entry so very quickly, within half a year everyone had done it all And once the AI model capabilities caught up it was basically all done

55:01So we happen to be in a period of transition between the old and the new The model's capabilities have already spilled over into many simple tasks that it can handle on its own But, on the other hand we haven't found a sufficiently big scenario where you can use the Agent approach to really let AI do something complex enough and valuable enough while making very cost-effective and full use of intelligence and also building your own unique technological accumulation So most people actually underestimated how hard Agents are Last year everyone was kind of like kids playing house, in my feeling at the time I have a new toy

55:39There are so many fun things And then everyone started jumping in and playing with all sorts of things No one thought about how big that thing could become Because the evaluation criteria back then were very simple As long as you had ARR, it was fine As for ARR, a new technology is of course valuable There's always someone willing to pay And then everyone had a good ARR and everyone was happy So by the end of last year my feeling across all of Silicon Valley was that it was an entrepreneur's paradise Everyone was having a great time there But this year it's like OK, the model's capabilities made a big breakthrough

56:11And that earlier batch of kids playing house got swept away But this year, perhaps about intelligence I've heard this common sentiment Everyone feels that intelligence can really do work now It can be trusted Because reasoning ability has reached a certain level And then people have really started to wonder can I seriously use it to do some important work That's the real sign that the application phase has begun So will model companies devour everything? You must have heard that claim too This year, models devour everything No. The premise of models devouring everything is that everything is within the model's reach A model developed under the current paradigm

56:51will definitely go through a stage of development where the intelligence is good enough And then at that stage a whole bunch of new technical requirements will emerge to build things for different scenarios Then it's impossible for the model to devour everything Let me give a very intuitive example It's a really good example In 1892 Edison invented the light bulb and lit up Manhattan within a one-kilometer range a one-square-kilometer area of land Everyone knows this Then in 1910 there was a substantial breakthrough (In the US) roughly after that power plants started to appear the concept of the power grid Around 1910 the grid roughly covered the United States

57:38probably the major cities And then in that year two big applications finally appeared It was only in 1910 that two big applications appeared Besides lighting electric lighting that was the first big application the two big to-C applications that really appeared in homes the washing machine and the refrigerator only appeared around 1910 And they also used electricity Right, there's another very important one the radio And then the third household appliance we can't live without actually didn't appear until 1925 That was the air conditioner We definitely can't live without air conditioning This year's scorching heat in Europe has already made many people realize the importance of air conditioning

58:25To this day we can't live without air conditioning The television only appeared in 1930 So the development of applications definitely follows its own rules And today we can't live without the washing machine, refrigerator, and air conditioner But the radio and the television have been compressed into the mobile phone These two are basically gone The television still has a tiny bit of presence But it's basically been replaced by the computer and the phone So what I want to say is at the company level there's a very interesting choice Edison invented electricity but he believed power generation was more important So he founded General Electric which initially made power generation equipment

59:05But he didn't realize that the power grid was very important So his assistant left and founded America's first power transmission company But GE's real success came from producing the earliest refrigerators and washing machines This is very interesting So it was actually because a to-C application broke through that the power grid became infrastructure an infrastructure for to-B services People no longer cared about the brand of the grid company But General Electric prospered for sixty years and built all kinds of applications step by step But what we just talked about were hardware to-C products There was no software back then Because there was no knowledge back then no digitized information

59:47Not until the Internet era No, it started accumulating back then too But that accumulation was very shallow IBM was founded in 1910 because these devices started to appear and data needed to be processed Modern factories emerged so people were needed to manage information But IBM's real big growth only started after the computer appeared So it depends on who can better use intelligence in the intelligent era to build a to-C whether to-C or to-B product and whether hardware or software Yes You see, today's application companies as they keep going start wanting to train models because they feel without their own model they might lack competitiveness when competing with model companies

1:00:25like Cursor So do you think application companies should train their own models? Let me clarify the concept again I think we should still treat the Agent or intelligent agent as the main actor of the new era The key to entrepreneurship the boundary between them and models will gradually become clear And as I also mentioned earlier the technological innovation of Agent companies could also be very significant It just may not follow the same trajectory as today's large model technology innovation And at the same time because large models compete with one another the capabilities large models can offer will become a commodity an infrastructure So as an agent company

1:01:08there's actually no need to develop such a large model anymore And as for Cursor it's not the kind of main actor we define It was just back then when coding capability wasn't good enough it was testing where the boundary of model capability lies So once the model evolved it actually got fold into absorbed right in becoming nourishment for model iteration So these are two completely different stages That's why over the past two years it wasn't easy to find that pattern It's because all these things were evolving dynamically But once you understand the fundamental difference between the development of infrastructure and that of Agents if you ask people working on Agents now

1:01:56they will probably believe that their future technology and large model technology should have a fairly clear boundary So Agent companies don't need to train their own models, right? No need to train models in the traditional sense Because you can reliably call them without worrying that a model company will cut you off Because there's a reasonable substitute that can plug in anytime Does Manus need to train its own model? For Manus, it depends on what kind of model it wants to train next Because when you build an Agent you also have to make big technology investments So Manus happens to be stuck in the middle It has to explore where that boundary is

1:02:35But these may all be problems we encounter at this stage like how model companies should build applications and how application companies should build models Gradually, the industry's ecosystem may become clearer Yes Once a few decent to-C Agents start running people will roughly get a feel for it Why did OpenClaw flare up for a moment and then the enthusiasm faded? Not really First of all it has a niche user base After all, it's still geeks and CEOs who are most interested It hasn't expanded to the general public Second the technology is indeed immature now Most CEOs are CEOs willing to try new things have all been tormented to the point of misery

1:03:13Very few can actually get the lobster up and running So, that said it served as a very good rehearsal showing everyone the potential and the direction But the product itself is an early-stage one It's what we call the enterprise development cycle or the innovation diffusion cycle It's the earliest one that geeks try out an experimental product Up to today can we say that a really good to-C application a to-C Agent hasn't appeared yet? At this stage within my limited scope I haven't seen one But it will appear It's not as pessimistic as everyone thinks Right? No I'm very optimistic It will definitely appear Are other people very pessimistic?

1:03:56This year has hit a historic low No That's because, from an investor's perspective he's now having been burned in the last round and in this round not seeing anything decent he may not want to look anymore Think about it, if there were no good to-C applications what do we need AI for? Is it possible that the applications in the end are actually built by a few model companies the biggest opportunities It's possible OpenAI has always wanted to do that Right But whether it can pull it off depends on its organizational capability its innovation capability So model companies will definitely want to do it and will definitely do it

1:04:25The question is whether they can succeed in the end That's a separate matter Is there anything you'd like to say to application company CEOs who are struggling hard right now the CEOs of application companies? Just find your own direction of development and do your own thing well But fundraising is a very real problem I understand that too There's no way around it In the early days when you're doing something completely new the early stage is just difficult In 2026 there's actually another phenomenon Besides this wave of application companies many neo labs have appeared Whether in the US or in China Their directions include AI for AI

1:04:58and world models These are the more common ones How do you view this the fundamental reason behind the emergence of neo labs? Why does this phenomenon occur? Why is there a need to explore new model paradigms? I think there are probably several aspects Some people may have seen a ceiling in the existing paradigm or, from first principles like Yann LeCun he believes from first principles that this approach is flawed or rather its ceiling is very limited So he finally has the chance to try out the ideas he has always had So this is one type That type indeed believes that perhaps the scaling law has a trend of diminishing returns

1:05:36a trend No matter how well it's developing now that trend will come sooner or later This is certainly a consensus among some people That's why everyone has come out together Third I think it's actually because the foundational capabilities of large models have reached a certain level so people can try out some things that differ from the current large models' technical path but are also very fundamental innovations For example, AI for science That's of a similar level of difficulty and also needs the support of large models But the things being done may differ from the current large models' technical path So these several major categories are collectively called new labs

1:06:15Will neo labs become a transitional phenomenon or a long-term organizational form? I think they are the early embryonic form of a new organizational form The early embryonic form of a new organizational form So they will still exist long-term right? Yes They will continue to evolve Because You see, why is it called a neo lab? It's because people are dissatisfied with companies It means companies are going to be obsolete Really? Definitely Because the company and the industrial era emerged together There were no companies before the industrial era Only handicraft workshops So the biggest institutional innovation of the industrial era was that finally there was the organizational form of the company

1:06:57And then modern management could be put into operation Only then could it support the company's large-scale expansion It's a supporting institution If the industrial era's many fundamental economic laws are being challenged its organizational form will surely be challenged too That's why people prefer to call it a lab But as for this lab thing why do I say it's a prototype Because it more closely resembles the state of a group of researchers working together But for it to become the future ordinary people might also work in this similar more AI-native way So there will be some new evolution The real first neo lab is OpenAI Or including Anthropic

1:07:40So this batch of companies should actually be called neo neo labs Because they still borrowed the organizational form of the neo lab But they're starting to do different things The essence of a neo lab should be a lab, not a company Right Do you think this will become the norm in the future What will we call that kind of organization in the future I think it's uncertain right now And probably won't be called a lab either You see, in my recent book I deliberately talked about "organization" from start to finish I intentionally avoided the word "company" It's precisely because I feel that this year everyone is talking about AI-native organizations

1:08:16Why use such a complicated three-word term It's because nobody knows what this thing is It hasn't been defined yet This organizational form is still in its embryonic stage Maybe five years from now we'll have a new word called the new organizational form But today we don't So everyone calls it an AI-native organization Neo lab for short So you think the company as a product of the Industrial Revolution might disappear The corporate system of hierarchical management will die out Will die out Yes But the division of labor and cooperation among humans will continue Organizations will always exist Because where there are people, there are organizations But which form the organization takes

1:08:59is a process of historical evolution Why does the AI revolution bring changes so drastic so drastic that it feels like it's gone to the bone Even the organizational form has already started from the company shifting toward other directions Why is the change so big You can understand it with two analogies The first is to at least compare it to the Industrial Revolution Right And the Industrial Revolution basically overturned all the things of the previous agricultural era all of them So if AI is a technological innovation that surpasses the Industrial Revolution then most things of the industrial era will also be eliminated Second, from a more fundamental perspective

1:09:39Human beings have always been proud of being the only intelligent animal on Earth And the typical hallmark of that is that we have language and can speak Now AI can also speak AI can understand all human knowledge AI also possesses intelligence So from this perspective the changes in our future world are beyond our imagination So many things from the past that we take for granted may in twenty years completely cease to exist This should be a high-probability event How should an AI-native organization be built That's an especially big topic Let's talk about the most fundamental logic Job positions may no longer exist In other words, the job position is

1:10:30the most basic operating unit of a company The most basic operating unit of a company So when we apply for a job, we apply for a position When a company wants to grow it first defines a position writes a job description then goes to recruit people And only then can the company grow And because there are positions there are reporting lines there are hierarchies there is corresponding compensation Everything is tied to the position The position is a basic unit of today's organization But in the AI era, the basic unit of an organization is the task It's the thing that needs to be solved So everyone revolves around tasks

1:11:07This organization is definitely completely different So if we shift from positions to tasks how should people be structured within this organization This is what everyone is exploring now The most interesting question That's a very good question So the interim product is what everyone talks about: AI must be embedded in workflows But this is just an interim product Because that workflow was designed for humans Right, this still feels like existing companies and organizations plus AI Right, so an AI-native organization defines the task from day one Then breaks down the task Then allocates the task All people and AI revolve around the task Silicon-based employees and carbon-based employees

1:11:48all revolve around the flow of tasks collaborating Beyond most people's imagination This takes some time to digest But it's a matter of task decomposition and completion Then what will people look for when job hunting in the future There will be no more positions You go and claim a task These big labs now all publish tasks And then you go claim one If you think you can do it, you take it And once you can deliver the result your work goes live The company just publishes tasks And you just go claim tasks It's a more network-like structure Yes Its hierarchies will be further reduced They won't exist

1:12:27They won't exist, because hierarchy exists because nobody knows who has the final say So you set up a hierarchy first And whoever ranks higher gets the final say It's not based on judgment as the standard Of course, the hierarchy itself is related to the accumulation of judgment But in principle, in any given situation whoever has the higher rank has the final say But in the future, whoever is more capable gets called upon So this structure of junior, middle and senior levels will be gone too Positions, middle level, senior level, junior level That's why it's called hierarchical management It really has this kind of vertical division of labor

1:13:00It has this chain of command So once this chain of command is broken future collaboration won't rely on commands anymore So the middle layer in the traditional sense will definitely disappear Then how do you motivate employees Because setting up this kind of hierarchy part of the reason is also to serve as an incentive ladder letting you climb the ranks within the organization Exactly So where will performance evaluation and people's drive come from within an organization Essentially, it's self-driven That's why in the past half year the most popular term in Silicon Valley is "high agency" Right Actually, what we call it in Chinese is subjective initiative You are the master of your own destiny

1:13:38Whether you've grown or not nobody knows better than you Right So if you've grown and you're happy you keep doing it If you haven't grown and you're unhappy you go somewhere else And your competitors seem to be no longer just your colleagues but also your AI colleagues Indeed The first very important thing is indeed to look inward That is, you have to find your own drive Second you need to have Why is it called subjective initiative You need to understand what kind of capabilities you want to cultivate Third you need to find the scenarios where this capability is best applied And then grow along with that scenario

1:14:14Fourth (a slip of the tongue) Because you took on this task and you delivered the result this whole thing is very transparent So you will definitely receive the corresponding reward Why are these top researchers' transfer fees so high It's because contributions are very transparent You see, when you interview them I notice the titles you put on screen now are all "top researcher" There aren't even titles anymore They don't even write things like VP anymore There's no need, because everyone knows their standing in the field So because this entire network has become more transparent that is, who is contributing and who is not contributing is very clear

1:14:47Both companies have a top 30 people a top 300 people a top 1,000 people Whether you're in there or not everyone knows So the neo lab can't really be considered an interim product It might be considered an early form of an organization of a new era Exactly Because the reason it's called a lab is that the main employees right now are researchers What researchers understand best is a lab They don't like companies and they don't like universities either But they like labs So it's called a neo lab But in the future, actually all creative people will exist within these new types of organizations By then people probably won't be limited to the label "lab"

1:15:31But it probably won't be called a company either It definitely won't be called a company It definitely won't be called a company So the company will disappear The company will die out There's nothing to regret about that There will be more fun things in the future Think about it, how many people are happy going to work at a company now Not happy Exactly So in the future you could have a happier way Isn't that the value that technological progress brings to humanity Will the one-person company become a direction of development No Why? The one-person company just means many of the company's functions collapse onto a single person

1:16:02One person can now do the work that used to take dozens of people But the amount of work one person can do is always limited So you must have an organization There must be division of labor There must be cooperation So there must be organizations So the one-person company is an interim product It just helps everyone understand that a single person with the help of AI can actually accomplish what used to take a whole company with all that management and coordination cost to achieve So everyone in the future will be a one-person company You'll be an all-around player And then within that organization you'll collaborate with other all-around players

1:16:40working together So with all these new organizational forms emerging what should traditional companies do What should they do Their companies are already established Transform, upgrade Become a new species How to transform Change your mindset, or change the people Change your mindset, or change the people Exactly That is, your own cognition must be upgraded first So that's why now people are talking about AI-native cognition and ideas and why that's valuable It's because you have to understand what the new species will look like At the very least, you have to understand how big the chasm is that you need to cross Only then will you make the effort

1:17:17Otherwise, most people following inertia will just do AI+ and keep adding all the way down But when AI-native companies really rise up you'll be in despair It's just that this transformation is indeed very difficult Because it is in itself against human nature My original understanding was that this wave of neo labs is actually still some forms from the stage of early exploration But perhaps maybe in the future, for application companies this early form also has reference value Today's neo labs are indeed all doing early-stage exploratory work And it's technology-oriented So it's very natural for them to be called neo labs But future companies will also be doing

1:17:58very challenging exploratory things It's just that technology is only one part of it It may not even be the most important driving part So what should that kind of organization be called What kind of analogy do they like A club Or something else Right now they like What people actually like is the feeling of a team It's a professional sports team If you are an NBA star team the star player among them then that is definitely the greatest achievement I think this group of people will probably find a new way to name themselves These neo labs and ways of collaborating with AI will probably continue Right

1:18:39Are the current neo labs in a bubble Most likely, yes But a bubble is a good thing That is to say An early-stage bubble is definitely a good thing Because only with great rewards will brave people step forward And everyone sees similar opportunities This is the value of the market mechanism Because you don't know You don't have a God's-eye view in advance knowing who is right and who is wrong So you can only let everyone trial and error And then the market says OK, I know this is the direction So I'm willing to invest this much resource to support everyone's trial and error And this round of

1:19:10neo labs being worth so much In other words being able to start at such high valuations is because everyone believes that at the very least their starting point is OpenAI and Anthropic So when your potential return is this high then the bet you're willing to place right now is certainly bigger So you think OpenAI and Anthropic from the perspective of being just an infrastructure provider then shouldn't neo labs not be worth this much money No, it's the same Because you still get within ten years a company that goes from zero to one trillion That return is enormously huge So from an investor's perspective All they care about is

1:19:49whether OpenAI and Anthropic, in the next three years, can go from a hundred million-plus to at least three hundred million, or ideally five hundred million. Otherwise their investment won't even have a risk buffer left. But people investing in neo labs are more hoping that one of them has the chance to eventually become the next OpenAI or Anthropic. Because in your book, you also place organizational issues before strategic issues. Why is that? I mean, why— since you are a strategy expert— why do you put organization ahead of strategy? Theoretically, shouldn't we have the strategy first, and then design the organization? That's what we've always done in the past.

1:20:25It goes back to that fundamental judgment: that is, although the phrase sounds a bit clichéd, I still think it needs to be said— for the industrial era, AI is a paradigm revolution. So what we're seeing are new species. The common traits of these new species matter more than discussing any single part of it in isolation. So the reason that over these past two years I've been paying more and more attention to organization is that you discover the kind of future strategy you want to generate doesn't match the existing organization. Only with a new organizational form can you generate a new strategy. The two are like yin and yang in harmony.

1:21:09Why would a new organizational form emerge? Because it faces a very fundamental challenge: decision-making is happening at ever higher frequency, with ever higher demands on quality. That's why new organizational forms arise. And within the new organization, the most important decisions are precisely strategic decisions. So these two things really can't be separated. You have to explore them together: how do I use new methods to make new strategy, and then this strategy helps me enter the next stage, giving me the chance to try more organizational innovation. For an AI native company, if you were advising its CEO, what matters most in building the organization? Building a native organization

1:21:49has inherent advantages— it can build this native organization. It really comes down to whether your understanding of AI is deep enough, whether you yourself are native enough. For example, several entrepreneurs I've met who have gone far enough in this regard have basically rewritten their company's internal operating system. And because they code so well and AI helps so much, it's basically all like what Li Zhifei described: "I locked myself at home for three days and basically wrote out that framework." It's basically a similar story. That is to say, because they have a deep understanding of the prototype of how an organization runs, and AI's biggest help is that it can quickly help you

1:22:31bring that prototype to life. So they basically each built a system of their own. But most people today will be content with Feishu's AI upgrade, feeling that's already good enough for them. And since I'm already on Feishu, my migration cost is high too, and I couldn't build anything new myself anyway. And to judge whether what others have built is reliable—I probably don't have that ability either. So right here, this roughly becomes a watershed. That is, among the people of truly AI-native organizations in the future, the earliest batch of CEOs will probably have personally rewritten their own way of running the organization. Because that is the neural network of the future organization—

1:23:16its neural network, how it's connected together. And the workflows we talk about are actually, in their analogy to neural networks, essentially the same thing. Your old ERP was a fixed, paved road. A fixed, paved road inevitably has all kinds of rigidity and limitations. It can't possibly be a real-time responsive network, right? But a neural network's adaptability far exceeds anything else. A neural network is naturally matched to its tasks. As for CEOs who already have companies now, what's your advice? If you want to keep competing at the very frontier, you really do have to make a much greater effort on the AI-native front. Of course, competition is all relative.

1:23:58That is, if you're racing against a rabbit, the speed required of you versus racing against a cheetah— the speed requirements are definitely different. So before talking strategy, you have to talk about organization first. Organization has become more important for the CEO. Actually, they are two sides of yin and yang— there's just no way to fully peel them apart, because both are things that grow out of the same new DNA. So as I just said, that is, whenever I talk strategy, I inevitably get to organization. For example, as we chat along, I'll ask: How many people does your company actually have? Who participates in strategy discussions now?

1:24:33Right? And that moves into the organization part. And when you talk organization, you'll definitely ask: What's the biggest challenge your organization faces now? Your organization exists to solve this problem. So does your organization have a chance of solving this problem? So these two things are very tightly bound together. Regarding your vision of the future organization, do you think it might differ from today's organizations, from this corporate form, or what? Besides its center shifting from position-centric to task-centric, are there other differences? Maybe there will be fewer people? Fewer people—that's for sure. Or at least, for a phase, becoming fewer is for sure. In the future there might be too many fun things

1:25:07for everyone to play together, but that's another— a third-stage matter. We're only talking about the second stage. In the second stage, because AI's leverage is big enough, while the demands on people are relatively high, those who can adapt to this new organization are, at the start, a relatively small group— very proactive, willing to try new things, and with very strong learning ability. So they will figure out a set of methods for collaborating with AI. And once this set of collaboration methods has been figured out, maybe it can be opened up to attract more people, who can participate in relatively simpler ways. So interesting. I feel we're at

1:25:47the very beginning of the intelligence revolution. You should try to think of it as big as possible, not as small. Don't think it's just an internet-scale change. You should think it might be bigger than the Industrial Revolution. Yes. So the future could be an earth-shaking change. Yes. And everything has only just begun. And the imagination for what comes next belongs to everyone. Because in the first stage, you see, all the teams, all the entrepreneurs, are centered on researchers and technical experts. That's because they need to make huge early technological breakthroughs. But once the technology becomes infrastructure, all kinds of product operations, connection with users, empathy—things like that—

1:26:28become extremely important. I saw that in your book, regarding researchers, why they're worth so much money, you gave a partial rationalization: because they interact with AI the most, they best understand where AI's boundaries lie, and they can interact with AI the best. Right now they are the people creating the most value. And the market now is more efficient than before— it can relatively quickly and keenly identify who is creating the most value. So they receive what the market believes they deserve—this price. Do you think product managers will still matter in the next era? Of course they matter. Or rather, product managers will become incredibly important. Because what product you make,

1:27:10what customers you serve, what value you create, what technology you use— these are all decisions you have to make. Where to invest your limited time, on which relatively most important thing— these are all for the product manager to judge. It's just that today, the big model companies we see, because they must be technology-led, so what you see is, they're basically researcher-led. But there's a particularly interesting phenomenon here. That is, let me put it a bit sharply— this batch of entrepreneurs has a big hurdle to get over: whether they have an open enough mind to understand how the business world works. I've found that many technical people

1:27:57can't accept a simple fact: my technology is so amazing, I've created so much value— why might my company not be worth much? Because of a basic law of economics: as long as there's homogeneous supply, you can't capture high profits— you're just a not-very-valuable company. And this is completely different from the entire training in models that technical people received before. So those few technical people must complete this transformation. As a founder, as a CEO, they must understand the basic laws of economics, the operating laws of the business world. Only then can they become a qualified CEO and take the company far enough. Do you think product managers or founders of the new era

1:28:43need a researcher background? Do they need a technical background? The next generation probably won't need it. This generation does. Right—the model generation definitely needs it. The next generation, driven by applications, won't. Right. And when you get to the embodied-AI generation, you can already see that embodied-AI companies basically all have co-founders, and basically one has a big-model background, and the other leans toward applications and business. It's because they ran into this problem from day one. Right—it seems a founding team today has more co-founders than before. This is also an increasingly salient feature of AI-native organizations. It's because every person is indispensable— otherwise you'd just use an agent instead.

1:29:24So there's a very important judgment in my book: in the future, the entire enterprise will increasingly resemble a partnership. There's a very interesting supplement to this. As we said earlier, the company was the industrial era's greatest institutional innovation, and it replaced the old handicraft workshops. But there's an institution that existed in parallel: the partnership. Like law firms, big consulting firms, and even investment banks, early on— they were actually all partnerships. The typical feature of a partnership is that each of the few people's roles matters— you can't do without any one of them. So in future organizations, you'll find the situation where everyone is a partner will become more and more common.

1:30:10Otherwise I simply won't hire you. If I hire you, it's precisely because your contribution is indispensable. Then you're actually a partner— except if you joined early you're called a co-founder, and if you joined later you're called a partner. This may also change enormously. So everyone has irreplaceability. Yes. And the headcount shrinks significantly too. We just need to be a partnership-style organization— a team. Right. And everyone is differentiated too, not homogeneous. And each person has to be stronger. Right, definitely complementary. And if it just overlaps with what you cover, then it can be done by a digital twin. How many people is about right? That relates directly to the task.

1:30:51So just like OpenAI— their headcount is rising fast too, because the problems they're solving are getting more and more complex. The entrepreneurs of this AI era— have their traits changed compared to the past? The change at this stage is very obvious: it's scientists starting companies. But it won't last into the next stage. In the next stage, I think the share of scientist-founders will drop sharply. You've also said before— excellent founders versus extraordinary founders— what do you think is the dividing line, the difference? Can you spot at a glance which is an extraordinary founder and which is an excellent founder? You can't tell. Why do I say the extraordinary is hard?

1:31:22The most important sentence I've said: extraordinariness is recognized after the fact. It's because you overcame negative feedback again and again. When no one believed in you, in the end you proved with enormous success that you were the era's earliest prophet. So this process of becoming extraordinary is one that gradually unfolds. Excellence is a growth process with continuous positive feedback. So these two paths are completely different. Most people are excellent people, because excellence fits the human mindset better. But proportional to the extraordinary is what I mentioned at the start— many people are crazy and fail. Truly extraordinary people, people with crazy ideas who can reach extraordinariness, are also a very low proportion.

1:32:09There can be many wild dreamers— that's easy. But not only thinking it right, but also being able to execute it—that's an extremely rare ability. And if you ask these people, they may not think they're any different either. This is what "Built to Last" (by Jim Collins) talks about. The exceptionally great companies he saw— the other book is "Good to Great," right? Both books were written by the same author. He actually says that when you really ask these people, they share an enormous trait: they all attribute their success to others, and they all believe they were just lucky. He collected interviews and public communications from these people.

1:32:50I think at first it sounds a bit counterintuitive, but if you truly feel it, that's probably how it has to be. Why? Because the cause is bigger than the person. Because your ego must be small to go that far. It's the era that made you. Indeed. You just happened to be in the right place, with the right people, doing the right thing. So if you're only out to prove yourself, to prove how great you are— that's the Level 4 leader in his book— at some point, the world will slap you in the face to prove you're actually not as amazing as you imagined. So only with a smaller ego is it possible.

1:33:28Yes. Altruism, empathy, and a gradually shrinking ego are probably indispensable factors on the path to greatness. Because if your ego is big, you'll always want to prove yourself, and you'll stay stuck in that positive-feedback loop. Yes. So you'll most likely be walking a generally correct path. Yes. Because you're in a hurry to prove yourself, you need short-term feedback. You're doing it for your own ego, so you won't truly perceive and understand the real world. Your processing of information, from the very start, will be easily influenced by your preconceived judgments. And what if it's the other path? What he's thinking is: how do I maximize this thing—

1:34:10or rather, how do I do this thing well? Then I'm willing to listen to more people. different opinions I believe everyone is excellent We are here to fulfill this shared dream Then the cohesion of the whole company becomes different This is what a truly mission-driven company means So if you ask me I would also divide companies into four levels Opportunity-driven companies Strategy-driven companies Vision-driven companies Mission-driven companies But this mission has to come from the heart Something you truly believe Something you fully believe in and embrace Not something you made up and wrote on the wall to motivate others Such people are very, very rare How do you tell

1:34:54whether it's a mission or a strategy Maybe only you yourself know Maybe you don't even know yourself Entrepreneurs often come to me to discuss strategy The first question I often ask usually leaves them baffled I often ask Ten years from now what kind of self would you be satisfied with This question best reflects what you expect of yourself right now What drives this stage of your journey But this question is itself a question about ego Only when you have an ego would you think about what you'll be like in ten years For example, some people will say I'm just especially interested in this thing I hope this thing can succeed

1:35:33In that case, the cause is already bigger than the person Some people say I'm just curious That's also more interesting than proving yourself Then some people will say I just want to build a ten-billion-dollar company That's a negative-score answer Right Because that's basically ego-driven It's about seeking external recognition So many of the things he does will be adjusted and changed based on that He's not internally driven He's actually externally driven He's heavily influenced by external measures So every round of valuation easily impacts him It's like handing in an exam paper every time Exactly What's the best answer you've ever heard What Teacher Ma (Jack Ma, founder of Alibaba) said

1:36:15before he started his business To make it easy to do business anywhere That was something he genuinely believed from the heart Was that a mission or a strategy for him That was a mission Have there been any other great answers I think Musk going to Mars is also a mission That's something that transcends the self Right He wants to bring humanity into the space age For entrepreneurs which views are non-consensus and which are noise How do you tell them apart This is also closely related to strategy-making Because you are making judgments judging signal versus noise One is to try to perceive things more comprehensively That is, to genuinely experience what changes are happening in the outside world

1:37:02Second, your thinking needs to go a bit deeper That way you can gradually tell apart the noise from the signal Many people don't really think things through That is, if your thinking isn't deep whatever is trendy at the moment you just follow the trend So it gets exhausting Because you're always chasing Chasing trends while others are already two steps ahead You didn't see that You only see the third step when it surfaces So if you're chasing, you've already lost from day one Second, maybe it's because your exposure is too narrow So you easily end up making a locally optimal judgment But you may lose a broader

1:37:45perspective Right A lot of this feels like mental models, the "how" at the mindset level Personally, I like to ponder things at the mindset level And theoretical stuff can actually end up People easily get boxed in by theories Because you only remember that one sentence You just memorize that sentence And then you might copy it mechanically which easily leads to all kinds of problems But if you've mastered the mindset you can adapt to specific situations as they arise Since you also said that strategy nowadays starts with the organization and strategy also changes in real time So does strategy still matter today You've also said before "Look ten years out, think three years, act on one year"

1:38:28Does that still work in the AI era Today strategy matters immensely Make one important wrong decision and you're out This elimination race could be measured in a matter of months So that's why I say the AI era needs a new strategy methodology It's because the importance of continuously making the right strategic decisions is extremely high The more uncertain things are the more you need strategic thinking ability If everything were certain then everyone would just compete on execution, right Right Second, actually "look ten years out, think three years, act on one year" is my vivid way of summarizing the strategy methodology for the AI era But most people only remembered

1:39:20"look ten years out, act on one year" Because that's what they're used to The industrial-era strategy methodology Because in the industrial era or rather, starting from the 1990s the methodology of many traditional companies was because in a relatively mature and slowly developing world looking ten years out was fairly clear Because you grow 5% or 8% a year and you roughly know what things will look like in ten years And there were no new players So you more or less knew In that situation planning for the current year was also easy It was basically linear deduction plus a linear extrapolation And strategy mattered so little that big companies all outsourced strategy

1:40:03to consulting firms like McKinsey So from the 1980s to around 2000 firms like McKinsey and BCG saw explosive growth Precisely because these companies thought strategy wasn't important and could all be outsourced to others So what McKinsey used to do was based on industry experience based on our strategy methodology I'll make a ten-year plan for you and then you just execute it year by year Both sides were happy But our problem today is looking one year out is far from enough You just can't do linear extrapolation Linear extrapolation could never have predicted Anthropic's 80x year-over-year revenue growth this year So it's because you're a vision-driven person and you've given up

1:40:48the linear-extrapolation way of thinking that you can possibly have a future But since the future is so uncertain looking ten years out doesn't help you much Because you still don't know where to start right now So you have to force yourself to think about a picture roughly three years out So the real creativity lies in looking three years out Looking three years out requires you to see two or three steps ahead What's my first milestone What's the second milestone What might the third milestone be Then the first important thing I do this year is meant to serve which milestone So you'll have a dynamic tension among the short, medium, and long term

1:41:35This tension leads the company to potentially make better strategic choices in the present So without this three-year outlook without this tension, strategy actually has no environment in which to happen It's like searching for the optimal solution in three-dimensional space versus finding an optimal solution on a two-dimensional plane They're completely different things How long a cycle should we use as the frequency for thinking about strategy Good question So "look ten years out, think three years, act on one year" is a vivid metaphor Its essence is to highlight the creativity of looking three years out, and the conflicts and tensions between short, medium, and long term the trade-offs

1:42:21Strategy is essentially about trade-offs The core is exactly these trade-offs But you could also look five years out think two years, act on half a year It varies by industry You could even look three years out think one year, act on one quarter This rhythm can vary depending on how the industry develops But you must have medium-term thinking You must see at least three steps ahead not just one step So the key is that medium-term milestone Last year there was a startup an application company, that asked me Their company was doing very well ARR was growing fast I asked him just one question I said You

1:43:06after two more years of growth then what What will the endgame of the industry look like Does your ARR have a ceiling How fast will that ceiling arrive So he immediately understood He went back and did the math and found the ceiling wasn't high So he hurried back to start working on some new things But when you look at it linearly my ARR is growing every month the annual growth rate looks great Especially ARR — it's a very illusory thing Not only is it a linear extrapolation it extrapolates the present 12 months forward It easily creates an illusion of growth And then it may suddenly hit the top

1:43:47and afterwards go straight into linear growth And what you're doing is actually something with very little room for imagination Right, something without much value So your short-term value once the model spills over no one is willing to pay for it anymore Or there's too much competition When competition first starts pricing is based on the customer value you create But once there are many competitors pricing is based on your marginal cost That's also something most people don't understand — a basic law of economics Whether short, medium, or long term what should we use as the metric That is, what could serve as a yardstick for us to gauge

1:44:21our own growth right now This metric is subjective That's why OKRs exist They were meant to solve the problem of when your goals can't be converted into KPI numbers everyone should coordinate through OKRs It's just that implementing OKRs demands too much of the organization So most companies either gave up or OKRs turned into distorted KPIs Why is it that, in fact, in the organization of the future people now use the word "alignment" more and more It's because, to a large extent, it's really about alignment Aligning our goals Aligning our judgments about the future Aligning the different cognitive models behind us This has become something more first-principles

1:45:08Will OKRs carry over into the AI era I think the basic idea of OKRs is right But we may find that or rather, only in the AI era can OKRs truly be implemented Because that organizational brain that AI-based organizational system can track natural-language expression And this neural network is essentially ubiquitous So there's a good chance it could give us feedback saying "You've all deviated from your OKRs And I've seen some more specific things not necessarily numbers but some trends that I can remind you of" So OKRs may truly require a new set of methods to be implemented So in the AI era how exactly should these companies

1:45:50make strategy Could you talk separately about model companies application companies robotics companies How should each of them approach strategy I think you can't separate them that way Different companies are in different scenarios and at different stages of development As I also wrote in my book there's the strategy-exploration phase the strategy-formation phase the strategy-maturity phase and the strategy-execution phase So rather than dividing by industry it's better to divide by development stage For example foundation model companies have basically entered the strategy-development phase Because their strategies have converged Everyone is relatively clear now But embodied-robotics companies are mostly still in the strategy-exploration phase Nobody knows which direction to converge on

1:46:32So at these different stages the methods of strategy differ in some ways But going back to the more fundamental question how is strategy-making in the AI era different from before I think the biggest difference is it used to be called strategic planning now it's called strategic generation I deliberately coined this term Because this term inherently reflects the AI era Everyone is now used to video generation text-to-image generation Nowadays people say greatness cannot be planned Right So it's really about generation So for a company when we talk about organizational design what you really need to build is a system an environment where strategic insights can emerge and become a strategy

1:47:18a system that continuously generates strategy In the end, strategy is just a natural decision At the right time the right people make the right decision and the company moves forward So the key is to build that generative system That means organization is crucial Right, that's why I had to talk about organization first That is, if you don't understand that the organization of the future has the capability to do this It's a neural network It's a workflow, a task flow It's a task-flow-oriented system Both humans and AI are plugged into this system And the core is emergence Just like with foundation models In effect, humans and AI together

1:47:54are training a small model of the organization So the core is whether insights no one had thought of before can emerge whether the collective intelligence is evolving That's the ultimate goal How do you build a good environment Besides what we just talked about a neural-network-like system this is why a new culture is needed You need a culture that matches an all-partner workforce That's why "context, not control" has become popular again Five years ago, when I talked about this the Netflix case or even the ByteDance case most people thought it was impressive but had nothing to do with them But today everyone knows that you need to give context

1:48:41Give enough context and it's that powerful It can make the right decisions So alignment is aligning goals and cognition And then giving enough or even having everyone share a complete context Long context Right, and then things will naturally emerge So both humans and AI need to share it Right Because humans can't process massive amounts of information the way AI can But humans have flashes of creativity and the ability to see connections that don't yet exist What do you think about the fact that this year people really dislike the "old-timers" — the anti-old-timer companies Do you think this is connected to the organizational changes brought by the AI era

1:49:24to the essence of this cultural shift Is there some relationship there Because the "old-timers" they talk about are really the organizational systems of the previous era's companies The hierarchical system some forms that emerged from evolution and got left behind Yes, this is really interesting The word "laodeng" is really interesting In the past, everyone respected the old and cared for the young Right So what was the core reason? Because the experience of the old was valuable Old norms were being carried on So you had to respect the old and care for the young That's how you could learn things, right? That's what passing things down means So why does the word "laodeng" exist?

1:50:00It's what you just said On the macro level the old set of things no longer seems to have much value Companies are under growing pressure No one wants to stay in a hierarchy As Yao Shunyu put it very bluntly Who wants to play along with the "laodeng" and sacrifice their own youth? So this is a very typical mindset But at the individual level the reason this can hold true is that experience, relatively speaking especially experience that can be turned into knowledge, is no longer valuable because it has all been absorbed by large models So these smart young people by making good use of AI leverage you can instantly become

1:50:39an expert in any field That's why AI can now reach the level of a chief physician That's because all experience has been distilled into large models So your 20 years of experience So your 20 years of experience what value does it still have? Of course young people no longer respect the elderly And then a somewhat negative label like this emerges So what should older people do? Either retire in peace or become young again So what should the culture of organizations in this new era look like? The core of the culture is creation It's like a sports team Everyone is indispensable But there are still stars There are forwards

1:51:21and defenders — a division of roles So its culture must be open and transparent and shared There's another very important one I call it co-creation That is, everyone understands that they are creating something new together So the willingness and ability to cooperate and collaborate will also rise significantly Why do you think the companies of the last era made people so unhappy with that system? Few people were happy, I suppose Well, there was no way around it Because most knowledge work is simple, boring repetitive and tedious with no room for growth So we naturally feel that it's just a job It's draining It brings neither emotional value

1:52:09nor growth value And to manage people, companies all resort to control-oriented methods Right? And they watch you very closely So that whole environment increasingly fails to match what people today aspire to in terms of civilization or self-development What is happiness an understanding of what joy is So I'm very optimistic myself I think AI in this era has given us greater freedom to pursue what we want that kind of better life So future organizations may be a bit happier and a bit more creative I think happiness may become a necessity Emotional value is a necessity So the requirements for founders have also changed Yes, they've changed You can't have a boss mentality

1:52:59Who would still defer to you, right? I won't even play along with the "laodeng" why would I play along with you, the boss? This is a completely different environment So you see, all founders today say their top priority is hiring When hiring people are drawn in by inspiration It's not about you saying you can't PUA them Right, people won't buy PUA either A service-oriented CEO Yes Just now we were discussing strategy We didn't finish that part You said, for example, that large model companies have entered the strategic development phase while robotics companies haven't converged yet They're still in an early embryonic stage So specifically how should companies at different stages

1:53:35formulate their strategies? In the strategic convergence phase you need to pursue organizational efficiency and execution efficiency And you need to pursue going from a directional strategy to an operating model that can close the loop Only then can it be efficient Because the core of the development phase is scaling You need to ramp up volume fast But the reason you can scale is that your prototype is already robust So it can be scaled The core of the strategic exploration phase is that the number-one leader must lead everyone to pursue different innovations And there must be a mechanism to bring down the cost of trial and error And gradually reach a consensus

1:54:20So at this stage you need to have your own convictions but you can't be too attached to them Because reality may prove you wrong You may find that someone else's direction is right At that point, it's about whether your ego can yield to reality Because you need to keep adjusting And everyone's starting point is constrained by their past training So what you can see is all path-dependent So now everyone creates together Can you quickly take others' best experience understand, digest, and absorb it and then turn it into the company's basis for the next round of innovation So in this strategic experimentation phase the core is that innovation efficiency must be high enough

1:55:06rather than execution efficiency Look at today's large model companies whether in the US or China everyone is doing Agentic Coding This seems to be a big consensus They have no non-consensus, right? Should they have non-consensus? Non-consensus is the neo labs Those who don't buy into this consensus go start their own companies And among the neo labs a certain counter-consensus may become the new consensus three years from now That's what's so interesting about the market It keeps renewing itself constantly evolving forward Existing labs will enter homogeneous competition It's homogeneous competition right now And that affects their commercial value Just as you said earlier So last week everyone was panicking

1:55:49Wasn't it fear of homogeneous competition? K3, right? Yes Because of this homogeneous competition it shattered the expectation that large model companies could enjoy high monopoly profits for a long time And because this profit expectation collapsed people started to wonder whether these companies could keep investing such high CapEx So the root question is whether these companies still have the strength and whether it's worth continuing to invest in infrastructure for them The infrastructure is all built to serve these large model companies If in the end the value they create is limited why would I invest in all this? So your current view is should we keep investing? First, I think large models

1:56:34still have huge room to develop So most likely infrastructure investment will continue But at some point there may be an inflection point where infrastructure investment due to cyclicality may exceed the real demand of large models But that from an industry development perspective may not be a bad thing Because application companies can then use the existing infrastructure at very low cost use the existing infrastructure So their barrier to entry is greatly lowered That is, invoking intelligence will become really cheap yet sufficiently powerful and stable Or in other words As long as the supply of intelligence hasn't reached the state everyone wants infrastructure investment will keep going So application companies can hang on a bit and wait

1:57:23But you can't just hang on and wait either Because if you're not in the game you can't accumulate the relevant experience When the big wave really comes you won't be able to catch it So application companies should still have hope Application companies should still have hope Don't be so discouraged Right, I do think there's definitely reason for hope And the reason they entered the arena is that they felt some kind of calling from the future We're just on the outside and may not feel their belief as intensely as they do What do you think about robotics companies whether in terms of models or embodiment since today they're still far from converging

1:57:55when do you think further convergence might occur How long might this industry cycle take to play out You just talked about these model companies moving to agent companies and that stage So what stage is robotics at now? I think robotics is still at a very early stage They're still figuring out how to land it in practice Because there are some core constraints such as data Unlike large model companies which had enough data when they rose So there are very practical constraints that keep them in a very early strategic exploration phase Because there are several approaches right now One is to start with the brain and pursue the generalization of the robot brain model

1:58:31Another is to start from say, a specific scenario and work on embodiment, on the body so that within that scenario it can close the whole loop through data And then this loop feeds back into its model capabilities Which side do you think should break through first? I think logically both paths are viable So you can't judge in advance which path will win Because the challenge of generalization is how big the space for generalization is Second, in a specific scenario especially a very large scenario whether your performance is good enough For companies taking the scenario route their challenge is the scenarios available now aren't big enough But whether their capabilities allow them

1:59:15to quickly expand into a big enough scenario to both do it well and have the potential for generalization So in principle, it's like climbing a mountain You can climb from the south slope or from the north slope It's hard to say in advance who will climb faster Can this stage of robotics be compared to some period of some industry? The simplest analogy is what they themselves believe they're still at the stage around 2020 for large model companies, before the ChatGPT moment Can it be compared with earlier industries from history? Cars That's a valid analogy too But with cars in the early days there were also thousands of companies

1:59:53And it's actually a good analogy For example, roughly from 1900 to 1920 there really were thousands of companies in the US all making cars Some people thought I have a mechanical capability I can knock together a car Anyway, the motors were all bought from others But then there emerged a native application namely Ford In 1913 Ford built the world's first electrically powered assembly line Modern manufacturing entered a new stage Only then could Ford, with extremely high efficiency produce a car that ordinary people could afford So that was the auto industry's first to-C breakthrough And Ford became an icon of the modern industrial era But there's an interesting episode later

2:00:44When Ford, through scaling and standardization selling only black cars became number one in America consumer demand began to diversify And then General Motors appeared competing with seven or eight models in the next stage It shows every strategy only works in the context of its own time It's not something that stays valid forever Right. Now it's like everyone can hand-build a robotics company But which robotics company can achieve mass production at scale nobody knows Maybe we're still waiting for that Model T moment Yes, we're waiting for the Model T moment The Model T was remarkable Being able to mass-produce at scale That's why everyone thinks whoever can first actually

2:01:23sell a solid ten thousand robots sell robots has already achieved something remarkable Is it possible that the robotics industry may develop at a pace more like cars rather than large models, I'm wondering If you want to make that analogy we might as well zoom out to what we just talked about after electricity appeared the emergence of electrical appliances Because based on electricity, there can be N kinds of appliances So if you ask whether the refrigerator or the washing machine is more important there's no good comparison They're not that different in size either, right? So we may end up with N different kinds of robots truly suited to different major scenarios

2:02:02Household robots are definitely one But within household use there may be a companionship-oriented type and a housework-oriented type And then industrial scenarios or all kinds of outdoor scenarios It's a ubiquitous thing Why is everyone so bullish on robots? Because it's a fundamental interface for AI to interact with the physical world So it's a bit like electricity Everything can be redone with electricity So I think your question is very good A better analogy for robots might be the development of electrical appliances That might be a better analogy And cars became the flagship among them Because tokens are analogous to electricity Right, cars were indeed the flagship But that's the outdoor scenario

2:02:43The indoor scenario is washing machines refrigerators household scenarios Right In the face of this huge wave which giants are safe? I've always believed that in the face of a huge wave no one is safe Whoever thinks they're safe is the least safe of all How unsafe are they? They should all be worried enough Their strategies are all AI first all in AI Is that enough? It's a necessary condition Whether it's enough depends on how disruptive the path of technological evolution is and on how capable new companies are of acquiring resources These are the main challenges to watch Do you think they, going all in on AI

2:03:22can still become companies of the next era? I mean, era-defining companies? The challenge is huge The challenge is huge Is there any basis in history for this? Companies of one era transitioning to the next and still doing very well are there many? It's not a question of how many it's a question of whether there are any Are there any? Basically none Basically none IBM did it once And Microsoft arguably did it once That is, Microsoft, to some extent missed mobile internet But it held on to the flame of search And it built up cloud computing So combined with its PC-era monopoly on PC operating systems So in the AI era

2:04:10Google got an early start and got another chance But for now Microsoft seems to be a bit unsure whether it can make it through Because its investment in models isn't enough It's actually already out of the game on models Although it invested early in OpenAI but now it can no longer afford to buy OpenAI Even Microsoft which has remained a leading company for a very long time its future may be uncertain What about Google? Google is like Microsoft in its previous chapter Maybe it can become a great AI cloud company But whether it can become a big consumer application that's really not so certain The entry point

2:04:58Right, gateway-level applications That's not so certain But because it has TPUs it has large models and all that I think Google surviving as a cloud provider shouldn't be a problem So this doesn't depend on your strategic choices it's hard for you to always stay at the table That's the despairing part That's not despairing at all People are born, age, get sick, and die Companies also have birth, aging, sickness, and death Nothing is more normal than that That's why new companies get a chance and young people get to have dreams What a wonderful thing If they monopolized forever until today, wouldn't the world be too boring Google won't necessarily be the Google of the next era

2:05:35Will ByteDance be the ByteDance of the next era? Likewise, most likely not ByteDance's models today are also pretty good At least in China they're in the first tier No—so ByteDance is very likely to become an AI cloud company including Seedance, which is done so well in video generation and multimodality it's already a world-class large-model company Because building infrastructure is easy you have that accumulation and foundation it has a certain inertia But the big consumer-facing application carries very great uncertainty Doubao, you mean? Doubao definitely isn't Otherwise they would definitely pour money into it regardless of cost The reason they stopped pouring is that Doubao isn't good enough

2:06:19not future-proof enough Doubao is more like a previous-generation mobile internet product No, Doubao is a first-generation AI product that is, the chatbot Doubao has no ability to take action The core of an agent is completing tasks independently Then couldn't they build agents into Doubao? That's exactly what they want to do That is Everyone wants to build the OS of the AI era So they definitely want to be the entry point that orchestrates the OS Can the entry point of the previous generation understand the next generation's entire ecosystem development and become the entry point of the next generation That challenge is enormous Because if you're not the de facto standard for agents

2:07:04it's actually very hard for you to become the entry point What you invoke are all previous-generation services But those previous-generation services may well be replaced by next-generation services I mean, they're going to be disrupted Will social relationships be restructured? Is Tencent safe? They will definitely be restructured Tencent, as the safest one why has it been so panicked recently Precisely because it might not be safe either Definitely restructured Social relationships relationships between people will all be redefined and the importance of agents in this will rise significantly So you're not optimistic about an existing product adding AI and agents to it You're not optimistic I think they're all transitional products

2:07:40Just like QQ didn't transition into the mobile internet era Or rather, QQ did transition but what truly carried the mobile internet era was WeChat Relationships between people will be restructured People's social relationships will be restructured Then what about the relationship between people and goods? They'll be restructured just the same I think Alibaba faces the same challenge They're working extremely hard Adding AI and agents won't work either? Then is it possible to build a new product themselves? To let the next era's product be born within my own company All companies will go all in on AI Just as I said back then during the mobile internet era

2:08:12everyone went all in on mobile internet once I've actually been thinking about a question For the PC internet, was mobile internet a disruptive innovation or a sustaining innovation? I think that determines the essence of the question My judgment at the time was that it was a "1.5" innovation It had strong disruptiveness but it also had strong continuity Because it's all based on the fundamental architecture of the internet with network effects and so on The technology also had this continuity But it also had its very disruptive aspects So you can see its continuity part allowed several leading internet companies to all make the transition going all in and successfully becoming mobile internet companies

2:09:03But its disruptive part in the end did indeed give rise to Douyin and Pinduoduo these two more native mobile internet companies And AI now? Overall, I'm more inclined to think that it's a disruptive technology It doesn't just disrupt the internet what it disrupts is the Industrial Revolution itself The whole thing is overturned So it's a revolution in the forces of production The internet was essentially a revolution in the relations of production So once everything is overturned the continuity part is not that strong So in this sense for existing species to adapt to the new species will be very difficult Your technology reserves may not be so linearly transferable

2:09:44So hoping that the next generation of products will also grow within my company is also very difficult Even if what I use isn't my current product and I use the next product to carry it forward letting it be born within my company that's also extremely difficult Is this due to organizational culture? Two aspects Organizational culture is certainly one of the constraints Because for a company of a hundred thousand people to transform into AI-native versus a ten-person company growing into a thousand-person AI-native company the difficulty is not the same That's a very important first point The second is it's about understanding platforms and ecosystems That is, if you haven't built

2:10:15the first truly successful agent you most likely don't understand how agents operate then you have no way to provide that infrastructure you can't become the de facto standard and thus can't become the entry point So you yourself have to, on agents truly figure everything out Only then will you have users and be able to attract more people attract more agent developers to develop on your platform So having a successful agent first is a prerequisite It has that black hole effect Yes You see, over the past two years xAI has also done many things Do you think their strategy was wrong? Strategically there may be problems too

2:10:55They don't have a particularly clear direction not as focused as Anthropic And I think organizationally there are definitely problems Let me share a detail Whenever I'm in Silicon Valley and come across an occasion with a group of engineers the person from xAI is always the most exhausted one They're the most overworked Right It's no longer overwork in the usual sense they're simply excessively burned out Perhaps the way of working with Elon (Musk) is also draining Although it may help you grow it can be especially draining So when you see them show up they barely have the energy to talk Innovation in any organization is actually just like a person

2:11:36When people are too tired, there's no innovation You have to be in a relatively open and relaxed state for new things to come out of you When an organization is wound that tight and in such a high-pressure environment to pull off a particularly big innovation is a very difficult thing And from what I've heard researchers say they don't think Elon truly understands large models So his original methodology that more hardware-oriented methodology of his actually, in the AI era he can't directly give the answers himself and he can't use a co-creative approach to let the organization provide answers Because he used to be all top-down Because he used to be all top-down

2:12:19Right? Yes He gave the first-principles answer and then pushed everyone to accomplish it in extraordinary ways But in this context he can't give that answer himself And he doesn't give the people below him enough space to come up with an answer I think xAI is definitely stuck on the organization So this is also what you said about the relationship between organization and strategy being handled poorly Yes He doesn't generate strategy through the organization It's all his own will Right What about Meta? Every time I feel Zuck is so young Why can't he keep up? He's young, but he started early and burned out early and burned out a lot

2:12:56Because generation after generation of these things are actually draining your cognitive potential And Meta has grown to such a size that internally allocating resources and such becomes extremely difficult And I think most researchers don't think Zuck understands models either So he can't truly make really important decisions Then the other two companies Take Sam (Altman) for example he really relied on Ilya (Sutskever) And Ilya could truly drive the whole thing And Sam is a great CEO So the two of them made a great pairing And Anthropic's overall organizational capability is very strong And Dario provides the mission and vision and grasps that aspect fairly accurately So Anthropic has shown

2:13:43the explosive power of a highly focused vision-driven company But OpenAI, because its foundation is deep enough so even as it turns around this round to chase Anthropic its speed is very fast But these two companies are essentially both startups of the new era So in terms of organizational culture and so on they're actually a better fit Then as for the other companies they each have their own problems What about Chinese companies? Do you have any observations? Chinese companies are actually similar Essentially similar Kimi, for example, might be a bit more native But Qwen, because it started relatively early and has the resource support of a big company

2:14:28it has been able to keep running forward And ByteDance—once Zhang Yiming himself realized that he needed to go all in and get into the game ByteDance's speed became very fast too Why does competition in China always converge so slowly? Is it because first-mover companies unlike American companies don't have a two-to-three-year head start? So they have a healthy period to develop their DNA And Chinese companies at least up to this round so far are still fast-following on American prototypes So when everyone starts at the same time it inevitably becomes a "war of a hundred regiments" Because what everyone sees is more or less the same picture Then they all go to Silicon Valley

2:15:18and go through all the researchers and everyone can poach one or two people So you end up with a situation relatively prone to homogeneous competition But within this even so those with genuinely deeper insight who may have started a year or half a year earlier like Yang Zhilin and Yan Junjie and also Zhipu these are actually more native not really the fast-follower type And clearly in this round of competition they're still running ahead Shouldn't people stop building large models now? Joining this kind of homogeneous competition again Large models in the traditional sense I don't think anyone is doing that anymore and no one would invest in it either

2:15:55Unless you're a new lab Xiaohongshu miHoYo Aren't they all doing it? No Their core is building applications They're all building large models No They build large models to fill in some basic capabilities so that when they actually build applications they can exert real strength This is different from the other two companies That's why you don't see their teams coming out much to show off benchmark numbers and such It's being stockpiled as a foundational capability for the future I don't think these two companies are willing to join the competition of large models as infrastructure as API factories But at the very least they believe the premise of building applications is

2:16:30that I must have a very strong model For companies building big applications in the next stage the boundary between model and application technology is unclear So if in this generation they fall too far behind they may not be able to catch the next generation Right So this is what we discussed earlier whether application companies should build models So it's a capability reserve But this matter is itself an open discussion They're watching the actual progress to see which stage it evolves to It's like this game of chess hasn't reached the endgame it's still in the opening layout phase For this year, whether it's models agents or robots

2:17:07do you have any predictions? By the end of this year, 2026 I never make short-term predictions I only offer some regularity-based understandings for everyone's reference What about, say, five years? In five years, I think the agent entry point will very likely have emerged That is, the second stage By then, roughly everyone will be able to see it clearly Anyway, in the mobile internet era, things like Douyin and products like Pinduoduo didn't appear until the third stage Right Like Google, Douyin Douyin and Pinduoduo, I think are all similar cases And Pinduoduo emerged very late Right So it might still be how many years away from us? That might be eight to ten years away

2:17:50That's such a long time But it gives application entrepreneurs a lot of hope Don't forget, OpenAI is already eleven years old Right? It's only in these past two years that we've felt this acceleration When artificial intelligence becomes the new means of production and infrastructure how much time will society need to digest this technological revolution? One generation One generation Twenty years at a minimum Because they are the natives Then if intelligence can be obtained externally like energy what do you think the differences between people will be in the future? I think it's the difference in creativity So Drucker had a very classic analytical framework He divided human society

2:18:27since the industrial era into three stages The first stage was the productivity revolution that is, assembly lines replaced handicraft workshops The second stage was the management revolution which began roughly a hundred years ago with Taylor's scientific management methods And then business schools appeared this new species The core purpose of business schools was to supply modern management with enough managers The third stage he called the knowledge revolution stage He observed that starting from the 1970s software became a very important direction of development And the core of software is processing human knowledge distilling knowledge and best practices down through the method of software into a standardized method of execution

2:19:19This is Drucker He passed away in 1999, I believe roughly then So what he saw was the entire changes of the past two centuries So borrowing his framework I myself believe that if viewed from the human perspective the AI era can be defined as the age of creativity That is, all the knowledge that can be processed in a structured way —that type of work— will be taken over by AI And on one hand, people are forced to develop their new potential But you could also say that people are finally liberated to develop their own potential And I think this space is very large And that direction is creativity

2:20:06the creative things that AI cannot do It's just that we need to have a better understanding of creativity and redefine it Perhaps creativity is not simply being able to write a song or paint a picture It may be more about the ability to originally define complex problems For example only after electricity was truly invented and assembly lines appeared did the concept of industrial workers emerge the concept of skilled workers and then came white-collar workers These were all things that people in the 19th century simply could not have imagined So I think The direction everyone should work toward in the future is developing their own creativity. And this creativity

2:20:50is a kind of original creativity. Right, of course. It's the ability to create something out of nothing. Machines already know everything that exists. It's the ability to create something out of nothing. Because if it's something that already exists, machines have already captured it as data. Right. This kind of creation, from the machine's point of view, might seem boring, because it could be all sorts of emotional companionship, the little things people want from each other, all kinds of things they desire, needs that were never developed before. In other words, maybe it doesn't contribute to great progress in the physical world, but it might be of great help

2:21:24to human growth and happiness. and happiness. So this is something where there could be many approaches we never thought of before. You just said that if companies will disappear, will business schools disappear too? Business schools are already struggling. They're all declining. Everything built around the company now, like offices, these visible things— will they all undergo huge changes? They all will, because the proportion of the virtual world will keep getting bigger. The proportion of the virtual world will keep getting bigger. Right. The metaverse direction wasn't wrong. It was probably just five years too early. So in the future, for humanity as a society, what kind of changes can you see coming?

2:22:08There will be earth-shaking changes. Some of the Gen-Z entrepreneurs I've met, who want to build consumer-facing apps— the world they imagine is completely different from the world we can understand today. That's really interesting. I think today's conversation has brought a very positive, optimistic energy about the future. I'm fairly optimistic. I've always felt there's huge room for development. Will it change the entire education system? If the demand for human capabilities changes— It definitely will. The old education system was essentially an IQ drill. Its essence was cramming knowledge. Developing intelligence came second. Cramming knowledge came first. That definitely won't work anymore. Think about it: from childhood on, how many of our interests were suppressed?

2:22:50And we never followed an open, exploratory path to develop our potential. From the earliest age, just so we wouldn't lose at the starting line, we were pressed into the fixed molds of the old world. So I think the biggest conflict ahead will be the conflict within the education system. Because young people actually already know that the future is completely different from what we imagine today. Ask any high school student, and they probably all know that with today's methods, they won't even find a job after college. They may have many ideas of their own, but we certainly don't dare let them go down their own paths. So they're stuck right there.

2:23:32But if you ask younger parents, "My newborn child— how should I educate them?" they definitely know they need new methods, they just don't know which methods to use. So this is a direction where our whole society really needs to invest significant resources and effort in exploring. The ability to create something out of nothing. Right. That's a great way to put it. Yes. AI has everything that exists, so humans need to create something out of nothing. Exactly. The relationship between humans and systems, between humans and AI— how will it evolve in the future? I think that within organizations, it will be very clearly a deeply collaborative relationship.

2:24:06And at the societal level? At least let's not talk about ASI— what artificial superintelligence will grow into. There's no consensus on that anyway. Let's set that aside for now. In this round, the AGI era that we can already see— it will definitely permeate every aspect of society. Just like over the past decade, we've become unable to live without our phones. In the future, everyone will wear a camera somewhere on their body, a voice-input device, and wearables all over their body, maybe even exoskeletons and the like. Humans will become enhanced versions of themselves. I think this is entirely imaginable. Have you seen any good hardware companies? Hardware products?

2:24:45Hardware products are actually one of the hot areas these past couple of years. All kinds of consumer hardware keep emerging one after another. As we said before, after 1900, everyone was making electrical appliances, household appliances. But what survived in the end were a few big categories, plus countless small categories. We don't know what those will be yet, just like the toaster that appeared very early in America, which everyone still uses today. That too was invented very early on. But it wasn't big enough, the technology wasn't complex enough, so it couldn't sustain a large company. So, there are many new companies now, and as we said, everything will be

2:25:19done all over again. But which are the big things, which are the small things, which are primary, which are secondary, and who, using what approach, will find the code to the future— all of this is uncertain. But I think one basic attitude needs to change. In the past, we tended to reduce uncertainty. But I think at this stage, you actually have to embrace uncertainty. Uncertainty is the biggest opportunity. So the mindset has to change first. You can't be afraid. If you see uncertainty, see complexity, and feel fear, you definitely won't be able to move forward. But if, facing huge uncertainty and complexity, you find it especially fun,

2:26:02and feel there's a special opportunity to create new things, I think that's the mindset of the younger generation of entrepreneurs. Maybe we've been worrying over nothing— we've worried too much on their behalf. Professor Zeng Ming, what is your own mission? I think it's to be a happy researcher. A happy researcher? Right, I think in today's context, that's quite an interesting positioning. A happy researcher. Will AI make humanity happier? At least that's what I hope. That's what people mean by "AI for good." But that said, it will definitely bring a series of new problems. But on the whole, on more philosophical levels, we could debate it. But setting philosophy aside,

2:26:46the progress of civilization is technology constantly bringing new possibilities, and then humans mastering the new technology and entering a new stage of civilizational development. Between the new civilization and the old, for human beings, is the happiness index going up or down? On a philosophical level, it's hard to say. But by ordinary measures, it's definitely progress. Just think about it: if there were no air conditioning now, how would you feel? And in the agricultural age, when basic subsistence was a problem for tens of thousands of years, the industrial age basically solved the problem of subsistence at scale. Then the consumer age solved the problem of insufficient consumption.

2:27:29Now there's basically oversupply everywhere. At our current stage, at least in physical products, we're basically in a phase of oversupply. Then the internet solved the problem of information asymmetry, solved the problem of letting everyone in the world share information. So that was another big step forward. And AI will certainly— once we have so many intelligent solutions, we'll be able to do more interesting things. So in the future, one of the criteria for judging an organization will be whether it can make people happier— the people inside it. That's been true since Google. People today may not have a feel for it, but Google in 2007, 2008 was like a god in Silicon Valley,

2:28:09even more influential than those two companies today. That's why so many people went to visit Google— just to have a meal in the cafeteria and soak in the atmosphere. And all the very best people wanted to go to Google. So back then, the essence of its culture was respecting people's differences, unleashing people's creativity, giving you 20% of your time to do whatever you wanted, providing you with the best cafeteria— right? So with that culture of theirs, Google was the first to sense, however dimly, what was coming in the future. So they were the first to propose —back then it was still called the internet age— that a new culture was needed,

2:28:44that new ways of organizing were needed. That trend will continue. I'm curious: what does a typical day look like for Professor Zeng Ming, as a happy researcher? Actually, first, I chat with entrepreneurs, to see their latest ideas. Second, I look at the latest information, watch interviews like yours. They really save me a lot of trouble. These days you can see very complete conversations of ideas, so I get a lot of information input for training my own "data," without tiring myself out too much. Otherwise, actually, I think in the first half of this year, most people have already deeply felt that you can very easily, because AI tools are so efficient,

2:29:35drive yourself into extreme exhaustion, because you simply can't keep up with its processing speed. So now people are starting to realize they need some boundaries, some insulation. And how to find, as a biological human, a reasonable rhythm, and then find a new balance with AI. I have a few rapid-fire questions left. A favorite food from anywhere in the world? We're doing this so that, before silicon-based life arrives, we leave behind more memories of carbon-based civilization. Stir-fried pork with chili peppers. A favorite place anywhere in the world? For now, probably still Shanghai. Do you usually read AI papers? What do you think are the important papers that shaped the course of AI?

2:30:14I don't read technical AI papers. Based on what you know today, what's one key, important bet? The emergence of a good enough consumer-facing Agent. Our company is called "Language Is the World" Studio. When you first heard that name, what came to mind? You might want to change it to "Agent Is the World." Can you share a few books? Your "books of a lifetime," ones that truly had a big impact on you? I think among the more foundational management books, there's the "Built to Last" series of books, and then Drucker's books. And there's one book from my undergrad years that I basically copied out in full, because I took so many excerpts from it

2:30:58that I practically copied the whole thing. It was Friedman's book about the fundamental operating mechanism of markets. Not a thick book, but absolutely brilliant. So that established my understanding of the basic operating mechanism of markets. What do you think made you develop such a deep interest in management studies, in strategy? Actually, back in high school, I gave up on the science track, on things like the Math Olympiad competitions, because we were probably the first class to split into arts and sciences, and I actually preferred things like history and the humanities. So even though at the time I was top of my grade and could have taken that path,

2:31:35I felt that for me, I just wasn't interested in it. So actually in high school, I gave it up. Then when I got to university, after choosing economics, I found that the economics we studied— the math requirements of economics exceeded those of most engineering majors. I found it too boring. So later on, when I did my PhD, I ended up choosing business school, and something more macro like strategy, a discipline that sits between the real world and the theoretical world and has to connect them. I found that more fun. Strategy is the brain of an organization, so it has to be done by the CEO. Do you think there's a connection between

2:32:09CEOs who are good at strategy and what kind of people they are? So among founders, not many actually have strategic capability. The ones who truly go far— as we said, there's mission-driven, vision-driven, strategy-driven. Reaching strategy-driven already makes you excellent. But not yet exceptional. The exceptional ones, strictly speaking, have to be mission-driven to possibly become exceptional. But some people start with a vision and gradually develop a mission. Others are naturally mission-driven. Or you form a truly rare team, where one or two people complement each other perfectly— one person might be stronger on strategy, the other stronger on execution. It's a bit like the combination of Wang Xing and Wang Huiwen.

2:32:51That's also a very rare combination. What kind of drive does ByteDance have? That's hard to say, because I haven't had much personal interaction with Zhang Yiming. So as for whether he has that mission part— whether it's there or not— I can't judge. But Zhang Yiming is a remarkable genius. He's someone who has his own management philosophy. And he's at least vision-driven, and he has a complete, personal understanding of organization, including his early idea that "the company is the product." That was already very ahead of its time. For example, when I used to talk about Netflix's culture, most people's first reaction was, "Chinese companies can't do that,"

2:33:31"because that culture is too" "American, too Silicon Valley." But ByteDance was the earliest to basically catch up with Netflix's culture— they're very, very similar. And in the end, ByteDance became an efficiency factory, because once it figured everything out, what remained in the strategy execution phase was the pursuit of efficiency. That's why its AI strategy required Zhang Yiming to step back in and redefine it. So in the end, every company ends up pursuing efficiency, because as your strategy gets clearer, all that's left is execution. But, as we said, the fate of organizations is that times will definitely change, making your old playbook no longer fit. So you either transform completely,

2:34:08or you get left behind by the times. It's fair to everyone. A new era. Right. That sounds like good news.

Where these words come from. This is the caption track YouTube holds for this video, uploaded by the creator. We read it, tidied the line breaks and laid it out so it can be read. The plain text version is at https://viewrankai.com/tools/youtube-transcript/E7qB_p1D0Xk.txt.

All rights in this video belong to Zhang Xiaojun Podcast. Watch it on YouTube. If this is your video and you would rather this page did not exist, tell us and we will remove it.