Those who listen to our Zero Shot podcast know by now that I have an aversion to black mirrors. I try to have one phoneless day every week, sometimes to my colleagues’ chagrin. I avoid music streaming platforms. And I generally limit the time I spend gluing my gaze to screens of any size.

But several friends (who also serve as great filters) sent me toward a curious trend on Tiktok and Instagram—Americans are making videos to say the current phase of their lives is “a very Chinese time”. There’s a bit of notional cosplay taking place here, and I suppose it’s just a bit of fun for everyone involved. As it happens, OpenAI is also living its version of a Chinese life.

Over in my stomping grounds, the reason OpenAI isn’t all that interesting to anyone in China—aside from the fact that it doesn’t (or can’t) offer its services in the country—is that its big moves aren’t novel in the eyes of local operators. A recent big swing, namely apps-within-an-app through its Apps SDK, is a “been there, done that” situation that elicits a collective shrug.

Some context: Tencent’s Wechat app is essential for people in China. It’s the first app that gets installed after anyone unboxes a new phone. Wechat is a combination of Whatsapp, Facebook, Instagram, Uber, the payments platform of your choice, every e-commerce portal, and every other app, period. It accomplishes this through “Mini Programs” that are precisely apps within an app, where the user doesn’t need to download multiple standalone applications.

This should sound familiar. Facebook Platform was a similar attempt in the mid and late 2000s. In the same vein, Google took a stab at it with Android Instant Apps in 2016, and held out with hope until last month, when the initiative was finally canned. The concept simply never really took off outside of East Asia, even though that sort of control is the holy grail. It worked for Tencent because Wechat already commanded 90% of China’s messaging market and had successfully integrated payments. So, making it possible to shop on a third-party platform without leaving Wechat, for instance, made sense. Storage limitations on cheap handsets and connectivity issues also contributed to the acceptance and eventual success of Mini Programs. 

Wechat’s creator, Allen Zhang, conceived Mini Programs as an extension of the social connections that the app created. During their launch in January 2017, he positioned Mini Programs through interpersonal interactions among people, and shared to strengthen social bonds. Two years later, he gave a marathon four-hour speech about his philosophy behind creating meaningful products, in which “social transmission” was a core theme. He specifically rejected the idea of making recommendations as a platform.

In other words, Zhang believes that social context shortens conversion paths. Instead of discovering an app, downloading it, creating an account, then inviting friends, Wechat users would share a Mini Program into an existing group chat where trust and interpersonal relationships had already been established.

You could, in theory, download nothing but Tencent’s Wechat app and be all set to order meals and groceries, book rides, hire house cleaners, renew your driver’s registration, file for divorce… and pay for all of it without ever leaving the app. The term “super app” wasn’t invented to describe Wechat, but it was the first to turn that abstract idea into a concrete reality.

Sure, sure, sure, you might be thinking, OpenAI’s moves rhyme with Tencent’s biggest success, but maybe OpenAI will be the one that takes it to a global level!

See what Cloudflare CEO Matthew Prince had to say at the World Economic Forum in Davos this week: “If the winner in AI is whoever has access to the most data, then I worry that Google’s going to run away with it… If you think about it, why was OpenAI started? It was originally because Sam [Altman] and Elon [Musk] were worried that Google was going to run away with the entire game, and today we’re seeing that Google is running away with the entire game.”

Tencent’s Mini Programs took off because it already had the entire country’s internet-using population on its platform. It won the social game and connected people with people, people with businesses, and people with services. Only then did it stretch another arm out to pull every other shred of online interaction onto its platform.

Don’t get me wrong—I really like the concept of OpenAI’s apps-within-an-app. If it works, this means less prompting, less wordsmithing, less feeling out the moods and vibes of a language model, and more precise interactions that give you exactly what you’re after. It’s the right choice to suppress the finicky nature of ChatGPT and take users to what they really want, when they want it.

But the world’s most popular AI product doesn’t have the prerequisites that made Mini Programs work for Wechat. More functionality is getting piled on, but it just feels like a desaturated version of the real thing—like notional cosplay.

Hi, this is Vidhatri. You might know me as the host of 90,000 Hours, one of The Ken’s podcasts. I’m now joining Brady, Praveen, and Rohin as the producer of the Zero Shot podcast, so you’ll see me chime in here as well. 

January 20, the day we recorded the latest episode (my first!), marked one year since Deepseek’s R1 model took the world by storm. Brady reminded us that Deepseek is not a typical AI company. It is the highly ambitious research arm of a hedge fund with a very reclusive founder who wants to get to artificial general intelligence (AGI). 

But what does that mean for AI research? How is Deepseek structured? And what is it doing differently? Brady breaks it all down—with all arrows pointing to where the AI race between China and the US is headed. Praveen and Rohin wonder if India has anything comparable to Deepseek. One lab comes slightly close—but there is a long, long way to go.

We discuss all this and much more, so tune in! You can listen to the episode on The Ken’s app, Spotify, Apple Podcasts, and Youtube.