Re-evaluating Tencent! "WeChat + Deepseek" - The first step of the AI Agent

Last night I wrote a WeChat + Deepseek first take, and some big shots inside Tencent kindly "criticized" it, saying that our perspective is too narrow, and the priority of monetizing search may not be the highest... but merely the first step in the layout of AI agents (connecting mini-programs/WeChat Work/WeChat stores, etc.)... I completely agree with this point about AI agents... After all, back in October last year, we discussed "the power shift in the era of AI agents" in our public account; at that time, Tencent was one of the AI agent entrances we were very optimistic about... When the news came out last night, I had a late-night chat with our own big boss... I have already expressed enough optimism about Tencent... However, pushing stocks in the secondary market always needs to be gradual; if we suddenly inject all the valuations of AI agents... the bosses/capital market may not be able to adapt... Goldman Sachs' trading desk observed this week that there is indeed a rotation from Alibaba to Tencent (which is also one of the most debated topics...); after this news from WeChat came out, this narrative should continue to ferment on Monday... Tencent's revaluation can basically be divided into three logics: The first is analogous to Meta's logic in the US stock market (which has risen for more than 10 consecutive days...), taking the route of cost reduction and efficiency enhancement; AI will greatly improve the monetization of various businesses within Tencent's existing ecosystem (including advertising recommendation efficiency, etc.); this logic has already been completed by Meta, making it easy to map and understand for the market. The second is analogous to the recent revaluation of Alibaba Cloud (Apple + Alibaba Cloud vs. WeChat + Deepseek), taking the route of cloud computing regaining growth; this type of valuation has already been done by sell-side analysts, so I won't reinvent the wheel; for example, Goldman Sachs' estimate of Tencent Cloud (currently, the market's valuation of Tencent Cloud may only account for 5%+ of Tencent's market value); this logic has been slightly reflected in the trading these days, and if some numbers can be realized at the earnings meeting, there should be considerable space. The third is the vast potential of AI agents, taking a "call option" route (analogous to the role of robotics business in Tesla's market value, needing to be valued based on a "possibility"); yesterday's news about Deepseek + WeChat started to add some fuel to this narrative; whether it can catch fire depends on the market's enthusiasm; this logic may take some time for the market to understand and imagine Today, taking advantage of the weekend, let's have a good chat about some thoughts on AI agents. Here we mainly discuss some frameworks, as for how much can be realized by Tencent in the future? We'll wait and see (after all, AI technology and business models are still rapidly evolving). Let's start the discussion with search. The logic of search is the logic of a "librarian." 1/ Since the launch of ChatGPT, discussions about "AI changing search" have been dominating the headlines (this is also why Google's valuation has been relatively low for some time); however, the effectiveness of using AI for search has not been very good. "Search" is the most intuitive imagination of AI for ordinary people, but during times of significant technological innovation, we often find it difficult to "imagine" something that has never existed before; Boss Wang's saying "the pit has not yet cooled, Shandong is in chaos, Liu Xiang originally did not read books" has always been very enlightening; 2/ I think most people do not spend too much time thinking about what search/internet is; we only care about the purpose of using the internet, which is (1) to obtain information + (2) to communicate with others; in the first stage of the internet, we organized all this information across hundreds of millions of "independent websites," and "search engines" (Baidu/Google) acted as "librarians" helping us navigate to these independent websites; 3/ With the maturation of the internet, especially in the mobile internet phase, we have established more targeted "bookshelves" for specific types of information. For example, when buying things, we search on Taobao; when booking flights, we search on Ctrip/QuNar; when needing life tips, we search on Xiaohongshu; as the "bookshelves" become more efficient, a smart and universal "librarian" starts to seem less important; in our local internet context, many young people have already begun to use Xiaohongshu for search, replacing Baidu search; 4/ Entering the era of AI agents, we begin to enter a "dark library" era (here I quote Bernstein, Mark Shmulik); in this era, we have a new way of obtaining information, which is no longer entering the so-called "library" (no longer needing to visit the vast majority of web pages) / "bookshelf" (not even needing to care whether your flight comes from QuNar or Ctrip, or whether the taxi driver comes from Didi or Uber), but rather standing at the library door, waiting for your AI agent to extract the information you need; indeed, there may come a day when we no longer need to open any web pages/apps to obtain information, so what does "search" even mean? 5/ WeChat's "chatting scenario" is precisely suitable to become the entry point for such AI agents. Imagine assigning tasks to your AI assistant through WeChat. The essence of the internet is to "reduce friction." 1/ In Internet 2.0, we relied on "librarians" like Google/Baidu to help us reduce friction and avoid getting lost among billions of independent websites; in the mobile internet era, we used specialized "bookshelves" like Xiaohongshu/Qunar/Taobao to reduce friction, browsing information in specific information groups based on our purposes; in the era of AI, we let AI agents help us reduce friction, focusing only on whether they can complete the "tasks" rather than where to obtain the information. 2/ How can AI agents help us reduce friction? This can be divided into three parts: (1) Information acquisition; for example, checking which hotel is better or when flight tickets are cheaper; (2) Personalized recommendations; for instance, when buying flight tickets, the AI agent needs to know our preferences for airlines, and when booking a restaurant, it needs to know what types of dishes we like; (3) Proactive actions; for example, if we've been using our computer for a long time, the AI needs to proactively ask if we're ready to replace it, or if a restaurant we like suddenly has availability, it should ask if we need to make a reservation. 3/ Completing all three stages may take a long time (development of technology + applications); however, the current level of technology can at least provide enough space for AI agents to perform in the "information acquisition" stage. 4/ One point to note is that the reduction of friction needs to be significant enough to lead to behavioral changes. For example, even today, some people still directly use Google/Baidu to find flight tickets (and then navigate to Ctrip/Qunar) or use Google/Baidu to ask for travel guides to the Terracotta Army in Xi'an (rather than necessarily searching for experience sharing on Xiaohongshu); for many people, such behavioral changes are gradual rather than overnight disruptions; even after AI agents are introduced, some people will still rely on old methods to obtain information until AI agents become good enough and the reduced friction is sufficiently obvious. 5/ Therefore, the next step is to see large models begin to perform tasks in a more personalized manner and have their own "memory." At that time, an AI agent that can remember your likes/preferences and help you avoid a bunch of complex checkouts/logins/searches/comparisons will bring about a qualitative change AI Agent What’s Next? I made many speculations about the power shift of AI Agents in my article last October, and many of those ideas are still applicable. Interested friends can take a look at: Power Transition in the Era of AI Agents (October 27) Here are some discussions I extracted from that time: A few thoughts: 1/ First, some conclusions; AI Agents will further shift power towards distribution channels like Apple (or WeChat); 2/ Back in May, I discussed the thoughts on Apple vs AI vs traffic; in the mobile internet era, Apple could draw 20% of profits from Google, and theoretically, the same could happen in the AI era; in the future, almost all software can be likened to traffic entry points like Google Search (different applications will need to bid); the way users interact with smartphones will shift from manually selecting applications (navigating and using applications on smartphones) to letting agents decide the "best" choice. 3/ This may create an opportunity where applications compete for user business. For example, a user might say, "Help me find a ride home," and different ride-sharing companies may want to win that business and bid for it. This conceptually is not much different from Google search results. 4/ Due to this competition across applications, the TAM (Total Addressable Market) for Apple/AI Agents will significantly expand; covering travel, food delivery, and ride-hailing. (This is the natural result of power consolidation); in the future, the TAM for Apple/AI Agents will not only be their own services. The TAM for all airlines, hotels, food delivery, and ride-hailing will become the TAM for Apple/AI Agents. BofA's rough speculation here is that by 2036, they might be able to skim 8% of the tolls; 5/ What does this mean? Who are the winners? Who are the losers? To put it in simple terms, it’s like this: Previously, you had to go to the market to buy food and drinks, call to book a hotel, and go to a 4S shop to buy a car; now, you suddenly hire a secretary (AI Agent), and the secretary, with your authorization, compares prices, finds vendors, and completes purchases. The role of the “secretary” / “butler” can generate a lot of power rent-seeking, which is self-evident in our local context (especially in the relationship-oriented society of my hometown). 6/ Who can play the role of this “secretary” / “butler”? (This was the thought back in October, and it seems more apparent now) Is it the developers of large models like Zhipu / Open AI? Or the hardware/device channels like Xiaomi / Apple? Or perhaps the ecosystem players like WeChat who have positioned themselves well (referencing the hard-core alliance vs game distribution example)? 7/ Changes in bidding methods Uber / Didi used to rely on the supply of drivers and the demand of passengers to conduct dynamic pricing; now, in the era of AI agents, cross-platform bidding will become possible (the bidding between Didi and Uber can be aggregated through AI agents). The pricing here is closer to the total demand vs total supply of the entire society; the space for rent-seeking may be lower (if the controlling party of the AI agent does not act maliciously, this model is more consumer-friendly to some extent). Of course, the profit margins for Uber and Didi will decrease. This also applies to Meituan / Ele.me, or Ctrip / Agoda. 8/ Platform Fragmentation Since Meituan is no longer the platform with the most supply relationships, as long as you have sufficient ground promotion capability / supply capability, you can capture a portion of the market share in the corresponding areas. We may see that Meituan has a higher market share in Beijing, while Ele.me has a higher market share in Shanghai (just an example); as long as they have better execution capability / supply capability in their respective fields. However, such execution capability will turn into physical labor, hard work. Previously, large airlines controlled the supply but did not control the demand (unable to reach customers), and this situation may also change (no longer needing Ctrip); KFC no longer needs to go through Ele.me. 9/ Big Data Reverse Price Discrimination Ordinary people used to lack computing power and time to conduct dynamic price comparisons. Now, AI agents can accomplish this (of course, you may need to pay a certain monthly subscription fee for the AI agent). 10/ Interaction between AI Agents and Agents; When an agent cannot dominate the market (for example, Huawei, Xiaomi, and Apple each have their own agents); how will pricing be generated? I haven't figured this out yet and need to continue observing. Changes in such ecosystems will further impact the shift of power in the era of AI agents. Recently, there have been many developments in the industry, and there are still some questions I haven't figured out; In the agent era, do companies that own data really have an advantage? If the agent is good enough to access Xiaohongshu's data like a human, will the data advantage within Xiaohongshu's ecosystem continue to exist? In the agent era, do we still need smartphones? If the friction in obtaining information and communicating with others is smoothed out by the agent, then do we only need a screen + camera (such as smart glasses)? In the agent era, with a significant reduction in friction, where will the remaining time of people go? Perhaps people will have more time for entertainment, thus short video/social systems will continue to be the winners. Here are some speculations from Bernstein that I think are well-written and worth referencing. Source: 180K Original Title "WeChat + Deepseek? The First Step of AI Agent..." Risk Warning and Disclaimer The market has risks, and investment requires caution. This article does not constitute personal investment advice and does not take into account the specific investment goals, financial conditions, or needs of individual users. Users should consider whether any opinions, views, or conclusions in this article are suitable for their specific circumstances. Investing based on this is at your own risk