Nvidia–Thinking Machines Deal, Tencent Enters China AI Agent Race, Vibe Coding Paradigm Shift

18:01Ken Brown:Yeah, so this is actually a great segue because Microsoft is being investigated by several antitrust regulators in Europe, in the UK, Australia, as well as the FTC here in the US. And specifically, those regulators are actually looking at the ways that Microsoft bundles its software and prices them, with the allegation that competitors have made being that Microsoft essentially, you know, uses its existing base of Windows and Office subscribers, which, you know, number in the hundreds of millions of paying users, and ropes those customers into using other products like AI or security tools or, you know, services on its Azure cloud by making it much cheaper to, you know, buy the bundle of those tools in tandem with the Office and Windows products that they already pay for.

18:52Ken Brown:So it's interesting that we're seeing Microsoft essentially, you know, create a new bundle that includes AI since, from my understanding, this is essentially what the FTC is currently probing as it decides whether or not to, you know, pursue an antitrust action against Microsoft. And where are we in that case so far? What's the timeline? Yeah, so I reported today that, you know, Microsoft has been asking competitors about Microsoft's alleged anti-competitive behavior in the last year. But especially, you know, in the last few months, I have heard that the FTC is essentially telling competitors to make sure that whatever they're submitting to the agency is like fully, you know, written out in a very clear packet.

19:38Ken Brown:And some people who are talking to the FTC interpret that to mean that the agency could be close to wrapping up its probe in the coming months. The next step after that would be for the FTC to vote, that the commissioners will vote on whether or not to actually pursue, you know, an antitrust lawsuit against Microsoft. It could go either way at this point. We don't know whether that lawsuit will come to pass, but people think that that could happen in the next few months.

20:04Venky Ganesan:And so the people that you identified today in the org chart, you put a little mini org chart together of the people at Microsoft that are leading the charge against the FTC in this probe. I wonder what was the common thread between all the people that you identified and what did you learn about how they're structuring their team?

20:22Ken Brown:Yeah, so Microsoft is very good at hiring policy experts who have experience in the government, often in the regulatory bodies that Microsoft is contending with. So, for example, Nick Banasevich, who leads the antitrust org at Microsoft, was previously one of the top tech regulators in Europe, who actually led some of the European Union's investigations, antitrust investigations into tech companies. They also have lawyers who previously worked at the FTC or at the DOJ doing tech antitrust action. And typically, those people obviously bring a deep understanding of how these investigations tend to play out and how Microsoft could best structure its arguments against allegations of anti-competitive behavior.

21:11Ken Brown:And so we're seeing people with those backgrounds being elevated to top legal roles within the company.

21:17Venky Ganesan:And Aaron, I just want to make sure I understand here, what is at stake here with this probe? We talked just now about the bundling that they are doing with AI, and this is a clear way that they're going to try to encourage people to adopt Copilot and really see the use of this software. I mean, this probe will still take months, maybe years to play out. I mean, what is at stake here? Would everything become unbundled? You know, what could we see?

21:46Ken Brown:I mean, it is definitely possible. We have seen in the past European regulators a couple years ago forced Microsoft to stop bundling its Teams messaging app with Office because previously, you know, whenever you bought Office, you would get Teams for free. Slack, which is now owned by Salesforce, lodged a complaint about that in Europe, and European regulators actually reached a settlement with Microsoft to stop bundling teams. So I think, you know, we probably wouldn't see anything that drastic come out of the FTC until after an antitrust lawsuit is filed. At the same time, you know, I understand the FTC is looking into other questions like how Microsoft sells security software.

22:30Ken Brown:So, you know, they may potentially pursue remedies and like that would dictate the pricing for that type of product. It could take a lot of different forms, but unbundling in the past has definitely been on the table. Great.

22:42Venky Ganesan:Well, Aaron, I want to thank you for coming on. That is Aaron Holmes, our Microsoft reporter here at The Information. Amazon is raising up to$42 billion in corporate bonds, according to Bloomberg. They are being marketed in up to 11 tranches that range from two years to 50 years. We have seen many big tech companies like Amazon issue these types of massive bond sales over the past few months, as they are all looking to fund their AI expansions. I want to bring on Ken Brown, our finance editor, to help us break down the news. Ken, what did you make of this Spawned offering that we saw this morning?

23:18Aditya Agarwal:Hi, Akash. It is, well, it's a lot of money, but Amazon has a lot of money. It's just more of a trend where these big AI developers are getting funding for the build-out, and a lot of it has been funded with their own cash, but they're all kind of stepping up and borrowing too this year.

23:36Venky Ganesan:How much appetite is there amongst bond investors for these big tech company issuances?

23:43Aditya Agarwal:There's plenty of appetite. You know, I mean, so, you know, Amazon has long sold debt because, you know, they built, they have a cloud business and a warehouse. They have all the warehouses. But a lot of these companies don't come to market too often like Meta or Microsoft. And so but either way, these guys have high credit rating. They generate a lot of cash. It's exactly what bond investors want. And so they've been gobbling this stuff up, you know, with little doubt.

24:13Venky Ganesan:How does that contrast with other parts of the debt market? I mean, is this a uniform story in every different asset class you look at?

24:20Aditya Agarwal:Well, you know, the stock market's been jittery. I mean, you know, we're in a war right now, right? And so I guess investors are seeing the bond market a little bit as a safe haven. You know, there's been some jitters there, but generally a safe haven. But the stock market's been pretty volatile. The commodities market's been very volatile. The other thing is the private credit market has been in a bit of turmoil in the last few months. And so that has been another place where companies have borrowed money for the AI build out. And so the private credit market has been hit by a lot of redemptions.

24:59Aditya Agarwal:People want their money back. They're worried about what's going to happen in the future. And so, you know, not that Amazon would need to go to the private credit market because that's more for lower rated borrowers. But, you know, that that has been a bit of a crack in the AI build out, you know, the funding scheme.

25:15Venky Ganesan:And so for people who who don't follow the debt markets as closely. So the hierarchy here is that the the companies that are better off, they go to bond investors and they have the luxury of of basically trading in and out of the in and out of these bonds. Private credit is a little more bespoke. You have to like really negotiate these deals. Is that the idea?

25:35Aditya Agarwal:Well, it's more like these are these are loans that go or credit that go to lower rated companies. The bond market tends you can go to the bond market, too. But, you know, the the private credit market has been a better deal for a lot of these companies. I mean, Amazon is a double a rated company, which is, you know, just one notch below the U.S. government. And so, you know, they they they don't have trouble. You know, there's no doubt they're going to pay the money back. So, you know, the bond investors just, you know, glom onto all this stuff. And I'm sure when the sale is done, which will be today, we'll know that it was multiple times oversubscribed and people wanted a lot more than they could get.

26:15Venky Ganesan:Now, we've seen other big tech companies go this route. Meta has done an issuance. I think Oracle, of course. who are the other names that you're watching other names you think could do these types of

26:29Aditya Agarwal:issuances well um one of the things you know as i said before the the bulk of the ai build-up has been funded by the cash that these big companies generate um so a lot of them now have said okay well we don't want to use all our cash we're going to borrow some and they get good rates and it makes sense you know uh there's a bunch of companies that could borrow more i mean meta could borrow more. Microsoft could borrow. There's no shortage of demand for them. And it's just a matter of them balancing their own corporate finance, financial situation. NVIDIA has been a borrower. They could borrow again.

27:06Aditya Agarwal:All these guys. I mean, the most interesting one was Oracle, which came out and borrowed earlier this year. There have been worries about them spending too much. But what they did was they borrowed$25 billion, but they also sold$25 billion in stock and like none of these big guys usually sell stock because their stock is you know so valuable um so it's much cheaper to borrow but oracle told the bond market essentially listen we're not going to go and hit you up for all this money we're going to keep our credit rating solid and we're going to actually dilute stock investors um to get this money so that was the interesting play none of the other big guys are going to have to do anything like that was it a was a smart play in your mind, in your opinion?

27:50Aditya Agarwal:Well, you know, it was a smart financial play. The underlying issue of how much they're spending on the build-out and are they taking on a huge amount of risk is another question. We have much less financial capability than these big guys and sort of go toe-to-toe with them could be a problem. And so they're seen as one of the riskier plays in the AI build-out. And it's, you know, so that's going to be the question. And I don't think there's a doubt that they can pay down the debt. I mean, they have a huge software business, a database business. But, you know, that's a more questionable one. The rest of the giants there, they could borrow all they want.

28:31Venky Ganesan:Great. Well, Ken, I want to thank you for coming on. I imagine we'll see more of these. And so I anticipate we'll have you on more and more. That is Ken Brown, our finance editor here at The Information. The race to develop AI agents is heating up. From Silicon Valley to China, companies are competing to launch AI assistants that can complete tasks for both workers and consumers. My colleagues in the Information's Asia Bureau, Channer Liu and Juro Osawa, reported exclusively this morning that Tencent is also looking to join that race, quietly building a new AI agent for its popular WeChat messaging app.

29:07Venky Ganesan:I want to play for you a conversation that our Asia Bureau Chief Jing Yang had with Juro about his reporting. Here is that conversation.

29:17Juro Osawa:Juro, could you break down for us what kind of AI agent is WeChat trying to build here?

29:23Jing Yang:So my colleague Chenner and I reported exclusively that Tencent's working on an AI agent that handles all kinds of tasks, including real-life tasks like grocery shopping or booking trips, for example. And in order to do this, Tencent's planning to connect this agent with WeChat's mini programs. And mini programs are like mini apps that run inside WeChat. And in China, there are millions of businesses that operate those WeChat mini programs. And WeChat users can access all of those services without leaving WeChat. So the idea is that if the agent can access all these different mini programs, they'll be able to perform all kinds of tasks, real-life tasks for the users.

30:12Juro Osawa:That does sound like it will have a lot of potential. However, I remember mini programs is not a new thing. WeChat launched mini programs around 2017. And Agentic AI has been the pursuit of many tech companies, both in China and in the US, in the at least last 12 to 18 months. So why is WeChat only doing this right now?

30:38Jing Yang:I think there are several different reasons, but I think the company has been quite conservative in its approach. And that's because WeChat is so important to them. and this is like a platform with more than 1.4 billion people. So that means pretty much everybody in China uses this and uses this every day. So if you launch something that's half-baked, that could have a big impact, negative impact too. So if you're going to launch something with AI for WeChat, something this big, it has to be perfect. So that really takes time. And also Tencent's been developing their own AI models, but people think that's still not very competitive compared to ByteDance, Alibaba, which are two of the leaders in AI development in China.

31:31Jing Yang:So Tencent's trying to catch up. They've been hiring for the team. But there's a question of whose AI model they're going to use for this AI agent and whether they're going to use the in-house models or they will use other companies' AI models.

31:51Juro Osawa:Right. But I guess if they use external models, maybe there are some unnecessary complication. It will be much harder when it comes to, in my mind, like data migration validate it with those third party models that kind of issues

32:10Jing Yang:yes potentially and so we've reported that they have tested uh you know third party models as well as in-house including so apart from this tencent model known as huynian hy they also have WeChat project. WeChat is separately developing its own AI models as well. And so there are different options. But we reported that they have tested different options so far.

32:42Juro Osawa:What are the options they have tested?

32:46Jing Yang:So, for example, they have tested open source models from Jipu, which is one of the upstart AI developers from Beijing that recently went public in Hong Kong, and also Alibaba model and DeepSeek model, and also the WeChat in-house model. And so, yes, and this project is still ongoing. So, you know, which model they will decide to use in the end and what exactly the end product will look like, it still takes time.

33:25Juro Osawa:It's quite interesting that Alibaba and Tencent, long-time nemesis, is actually potentially will be working together on this, or maybe more accurately, that Tencent would be considering using the technology developed by one of its biggest rivals. I think that really maybe is a comment on sort of how not great the HUI or HONYUA models are by comparison?

33:55Jing Yang:Well, we have reported also that Tencent has been really trying to build a stronger team for its own AI development. And so last year in September, they hired a researcher from OpenAI and he's now leading their in-house AI development effort. And so they definitely are trying to step up in that area as well.

34:22Juro Osawa:Right, I think it's gonna get really heated. But speaking of Alibaba, just circle back on that part a little bit. You also reported recently that Alibaba has a Qn app, not to be mixed up with the Qn series of models, but the Qn app has also started developing a lot of agente commerce features, you know, integrating with e-commerce to travel booking to payment to navigation, all kinds of business units within Alibaba's umbrella. What do you think this move coupled with what WeChat is doing is going to shape the competition landscape going forward for AI agents, for consumers in China?

35:07Jing Yang:Well, I think this year, the battle over AI agents will be the most critical battle in China's tech industry. And as you mentioned, Alibaba has been really doing a lot in this area. And they consider and turn their AI app into a full-on AI agent a top priority. So that's why they integrated a lot of the main services, including e-commerce, into their AI app. So they are definitely pitching, especially during this Chinese New Year holiday last month. Alibaba ran a big campaign spending a lot of money giving away e-commerce vouchers for people to try AI e-commerce through the AI agent. So they are definitely stepping up.

36:02Jing Yang:And in some ways, they are ahead of Tencent. And so is ByteDance. because ByteDance has also been doing a lot with their AI app. And ByteDance's Doubao AI app is the most popular AI app in China with about 300 million users last month.

36:23Juro Osawa:It's going to get really intense, I have a feeling. Thank you, Juro. Thank you.

36:30Venky Ganesan:That was Juro Osawa and Jing Yang from the Informations Asia Bureau. Vibe coding has caused many folks in Silicon Valley to rethink how they approach software development altogether. And Aditya Agarwal, a general partner at South Park Commons, recently wrote an opinion piece for the information on what this shift means to him personally. Aditya was previously the CTO of Dropbox and an early engineer at Facebook, and he wrote about how this initial sadness that he had about the transformation in coding has recently turned into reckless energy and eventually mastery. I want to bring on Aditya to help us understand how he is thinking about this paradigm shift in engineering.

37:10Venky Ganesan:Aditya, welcome to the show. It's great to have you here. Hi, Nikash. Pleasure to be here. Thank you for having me on. So tell me, why did you decide to write this piece?

37:19Aaron Holmes:The reason is because I think it's very easy to have a point of view that AI is essentially like a doomer prophecy or AI is essentially a everything will be completely fine and we'll all just be very happy and satisfied. But not enough people were writing about what it means to go through the process of having something that you are very good at to now kind of essentially be capable of being produced by AI. And it was a really interesting mix of, I would say, personal emotions and feelings that I went through. So I just penned something that came from the heart. And I think it kind of struck a chord with many people who are, I think a lot of us are going through that same kind of process over the last, call it six to seven weeks.

38:08Aaron Holmes:So what is it that you're feeling?

38:11Venky Ganesan:You've been an engineer your whole career. You have built these programs from the ground up. You've worked at the highest levels of these tech companies. What was it like for you to see these agents and these vibe coding tools do it for you?

38:26Aaron Holmes:I think it was, it's actually just profound wonder to kind of start, you know, because I think that, you know, ChatGPT came out roughly about three and a half years ago. Coding agents came about seriously, maybe about 18 months ago. And I think in that time period, every time I tried it out, you know, I could use it and I, you know, I would have this mental kind of reaction like, oh, this is cool. It can do all the simple stuff, but it can't really make all of the things that I'm good at making, the complex interplay of data and compute and storing data correctly, making it efficient. And to go from that sense of the AI is a good junior programmer slash a good intern kind of programmer or somebody that can assist me, to now being able to one-shot end-to-end produce things of high complexity, it was just a sense of profound like wonder that it could actually happen um and once that happens the second reaction that i had was actually a sense of grief which is that this thing that i was world-class at this thing that i took a lot of pride in essentially being a part of my identity even though i mean i i'm an investor today i don't write code you know the part of me that always felt as though i'm still a world-class engineer i can kind of you know essentially uh hang it with the best.

39:42Aaron Holmes:But to see AI now being able to essentially be at the same level of kind of what I can do, there was grief, there was sadness, there's a sense that something is being taken away from you. And that was the first initial reaction. And I sat, I think, with that grief for a couple of days before you kind of realize that, oh, if AI can do the things that I used to do, then what can I do now? And I think that it's replaced by a sense of just a sense of like manic energy that every moment that I'm not actually essentially producing artifacts with code to automate parts of my job at South Park, automate parts of my personal life, it feels like almost wasted time because these tools are so capable and powerful.

40:28Venky Ganesan:You know, you wrote this line in your piece, You said, if anything, this shift is teaching me what it is like to be human again. What did you mean by that?

40:37Aaron Holmes:I think that part of, I think, being in Silicon Valley in particular, and part of, I think, evolving your own, I would say, journey, your career, your life, is that I think there come these moments where you realize that the only option that you have is to essentially take a step up and essentially retrain and kind of in some ways ready yourself for the next part of your life. And I think this kind of change happens to us as humans pretty often as we go through high school to college to your first job to your second job and as we all kind of transform. And it was just a sense of, wow, the AI is in some ways forcing my hand to really introspect about what it means for me to be me.

41:21Aaron Holmes:And I think that part of finding your own personal identity over and over again is a deeply human kind of emotion. And I hadn't quite felt that in such a powerful way for a long time, powerful and visceral way for a long time.

41:36Venky Ganesan:So the way that I understood what you're saying here, and correct me if I'm wrong, I mean, I sort of read it as someone who is so senior in their career progression. And like we said, You were the CTO of Dropbox at one point. I mean, what it means to be human, I think, at its core is learning new skills and grappling with how you learn. Look, we're talking about vibe coding today. I was learning how to play the piano again a couple months ago, right? And it's that repetitive motion of getting frustrated, trying it again. And it sounds like to me, I mean, I hear your point around being human, but for me, it's just learning a new skill that you have to, But even if you're 60 years old, I mean, you have to do it now.

42:22Aaron Holmes:Yeah, it's a good, I like the way you phrased it, which is it is about learning new skills. But I think part of learning that new skill is confronting in a way that you might, you know, in some ways you have to go back to a beginner's mind. You have to kind of go back and admit to yourself, you know, I've been in Silicon Valley for 23 years. None of that shit matters, right? Like in some ways I can lean on my experience, but I'm also starting at step one in some ways. So I think it is a reckoning with the fact that if I want to learn this new thing, I have to accept that my experience counts for a bunch, but it also doesn't essentially immediately allow me to do everything that I thought I could do in the past.

43:03Aaron Holmes:How is it changing the way that you're hiring? Oh, it's a great question. um i'd say that uh my favorite you know i'm giving this away i guess you know this is kind of uh my interview questions but my most important question today is how have you used ai in your daily life what did you use to essentially make your make your work life your personal life better this week what have you built for yourself and a lot of a lot of kind of the underlying motivation there is that it's it's kind of weird i've noticed that there's some people who are running towards this new technology and in some ways embracing the change, embracing what it means to learn something new to be human again.

43:44Aaron Holmes:And some people are really digging in their heels. And my initial reaction was that there would be some correlator there in terms of perhaps, hey, listen, the more senior people are more set in their way. So they want to keep on doing things the way they have. The more junior people, in some ways, they're free to learn new things. But what I found is that experience and age, these are in some ways uncorrelated variables. So what I'm really looking for now is someone's comfort level and adaptability and kind of a desire to use these tools to essentially reshape who they are and what the job is.

44:18Venky Ganesan:So in other words, it's a bit of an oversimplification to say, oh, the younger generation that is coming up, people who haven't learned, quote unquote, the older ways of software engineering, it's an oversimplification to say they are at an advantage and that people who are later in their careers, you know, they have a little bit more inertia to sort of fight up against. What you're saying is it's a mindset and that you're looking for people at any point in their career who have really dug into this tool and say, you know what? I've been in Silicon Valley for 23 years.

44:51Aaron Holmes:I'm ready to go. I'm ready to go, right? Yeah.

44:54Venky Ganesan:But let me ask you this then. So how do you think that the curriculums in school compare this to when you were learning software engineering 25 years ago? I mean, have you talked to people, professors, about how they're adjusting their curriculums and making it a little bit more application-based or experience-based rather than just theoretical?

45:15Aaron Holmes:It's a good question. I would actually say I've been thinking a lot about this. I don't have a very concrete point of view on this yet. I would venture that this might be counterintuitive. I actually think that learning the theory, at least in my profession, computer science, right? I think understanding the theory, the math kind of behind how a lot of these things work is still very important because it allows you to reason through what these underlying systems and models are capable of. But at the same time, I also think a lot of the small toy projects that I used to do in school that are basically wiring things up, that's dumb.

45:47Aaron Holmes:Like the AI can do that, right? So in some ways, you want a barbell approach where I think you want to read in some ways the classics, like read the theory, understand the math, understand kind of like the theoretical underpinnings of, you know, programming languages, compilers, systems, databases. And on the other side of the barbell, you should be incredibly, I would say, prolific in terms of what you are producing with that theory, right? And I think the middle ground is kind of where a lot of colleges are stuck today, which is that they're teaching you how to do things without a solid underpinning of the theory.

46:17Aaron Holmes:They're kind of teaching you the craft, but the craft as defined five years ago, these agents can do today. So I push kind of the curriculum to be more barbell. I don't think enough people are doing it today.

46:29Venky Ganesan:So when people say that,

46:31Aaron Holmes:oh, you don't even have to learn how to code today, you actually don't believe that entirely. Well, I would rephrase it. I don't think perhaps you need to learn, quote unquote, how to code, but I do think you need to learn how the systems work, right? You should be able to, if you are a great engineer, even if AI does most of the writing of the code itself, you should be able to describe exactly what's happening in terms of how the data is stored, how the data flows, where the compute comes, where the constraints are, where the long pole is. But you don't need to write the code, right, in some ways.

47:03Aaron Holmes:And then in some ways, understanding the theory and the systems allows you to be a much better, I would say, conductor of essentially the orchestra of your agents, if you will.

47:12Venky Ganesan:Right. Well, Aditya, I want to thank you for coming on. It was a great piece. That is Aditya Agarwal, a general partner at South Park Commons here on TI TV. That does it for today's show. Reminder, we are on this stream Monday through Friday at 10 a.m. Pacific, 1 p.m. Eastern. I want to thank you all for tuning in. We really do appreciate your viewership. Make sure to subscribe to the information on YouTube and follow us on X, Instagram, TikTok, and check us out wherever you get your podcasts. I am already excited for our next show tomorrow. Have a great rest of your Tuesday. Bye-bye for now.