OpenAI is in the midst of overhauling ChatGPT. The goal is to transform the chatbot’s simple interface into a personalized AI agent that can handle tasks in every facet of your personal and professional life. The company has taken to calling this new product, privately and publicly, a “super app.”
The all-in-one platform represents one of the biggest bets OpenAI has ever made, and one engineering leader now holds enormous sway over whether it pays off: Thibault Sottiaux. Last month, Sottiaux was appointed OpenAI’s head of core products, overseeing both ChatGPT and Codex, as well as combining them into the future super app.
To make the super app a reality, OpenAI has already shuttered several of its stand-alone products, including its video app Sora and an AI platform for scientists. Many of the executives who led those teams have since left the company, while Sottiaux’s influence inside OpenAI has continued to grow. He now reports directly to Greg Brockman, who is currently responsible for all of OpenAI’s product teams while Fidji Simo, the company’s CEO of AGI deployment, is on medical leave.
Sottiaux was instrumental in helping build out Codex, which has become one of OpenAI’s fastest-growing revenue streams. But leading Codex meant serving developers and working with AI researchers. Now, he’s being tasked with something new: revamping a consumer product with nearly a billion weekly active users.
“It’s incredibly exciting and mildly terrifying at the same time,” Sottiaux said in an interview earlier this week.
OpenAI has started talking publicly and often about its plans to build a super app, but it’s still unclear what exactly the final product will do. The term “super app” is usually used to describe platforms in Asia like WeChat, which bundle everything from messaging to payments and shopping into a single interface. But OpenAI is planning something seemingly far more ambitious.
Sottiaux says the goal is to build the “world’s best personal agent that deeply understands what humans care about.” Over the next year, he says that ChatGPT will become “delightfully proactive,” and bring people the right information at exactly the right time.
OpenAI is hoping that turning ChatGPT into a super app revitalizes the company’s growth as it races toward an IPO and tries to ward off intense competition from Google and Anthropic. OpenAI's bet is that creating one personalized assistant for everything will make it the unequivocal leader in consumer, enterprise, and the overall AI race once again.
Sottiaux grew up in Belgium and studied applied mathematics before joining Google’s London offices in 2015, where he worked on Google Maps before moving over to Google DeepMind. There, he helped build the infrastructure and tools researchers used to build things like AlphaGo, which made history in 2016 when it became the first AI to defeat a human Go champion.
When ChatGPT launched in 2022, Sottiaux said he felt inspired to move to San Francisco and find a way to work for OpenAI. “This is something that we had been sitting on at DeepMind for almost two years, and we were just not doing it,” he explains.
Sottiaux officially joined OpenAI in 2024 and initially focused on developing tools for the company’s own researchers, just as he had at DeepMind. But within a few months, he started building what would eventually become Codex. As the AI coding tool exploded in popularity, Sottiaux became a minor celebrity in the developer community, personally responding to bug reports on X, and occasionally granting engineers’ pleas for him to reset their weekly token limits.
But in his new role as head of OpenAI’s core product, Sottiaux will be tasked with thinking about what the average person wants from AI—not just the needs of his fellow engineers.
In practice, I’m expecting OpenAI’s super app to be a digital assistant with advanced memory capabilities. It will likely be capable of, say, making dinner reservations, but also reminding you later to avoid menu items that contain allergens, or that upset your stomach last time. The platform could also help automate work tasks, such as filing expense reports before they’re due.
Under the hood, Sottiaux says the super app will largely be powered by Codex, which is already seeing strong growth with nontechnical users. To complete a task, the agent may write software code, run an API call, or surf the web, but the user won’t see any of it. They’ll just ask for things in natural language—or at least, that’s how it’s supposed to work.
Sottiaux says building the super app mostly involves converting Codex into a general-purpose agent, and then merging that system into ChatGPT. As OpenAI shuttered other initiatives, Sottiaux says the project gained additional resources, though the core team remains relatively small. He declined to say how many people are working on the super app now, but his Codex team consisted of only around 40 people two months ago.
This isn’t OpenAI’s first attempt at turning ChatGPT into an agent. Last year, the company launched Operator, a tool within ChatGPT that tried to navigate the web on a user’s behalf. It eventually morphed into ChatGPT Agent, but neither product ever saw significant adoption. Sottiaux says those attempts were “too early”—the models powering them weren’t reliable enough, so OpenAI had to heavily restrict what they could do. Now, he claims, the technology is there.
Another problem with OpenAI’s earlier agents was that consumers didn’t really know what to do with them. While software engineers have proven adept at using agents to automate a wide range of tasks, teaching people how to use ChatGPT in new ways will likely be a big part of the challenge Sottiaux is facing.
“We have to bring the user along. Initially, maybe it’s a small thing that we can do for you, and then increasingly, build confidence that ChatGPT can do bigger and bigger things,” says Sottiaux. “Maybe then you start teaching your peers, your friends, and your family these new capabilities that you found in ChatGPT. Then also the model in ChatGPT itself has a role to play there, almost as a mentor.”
Sottiaux wouldn’t say when the super app is coming, beyond “soon.” But he notes that “a lot of what is going to be made available for everyone in ChatGPT is already available in the Codex app,” and OpenAI has already said that it plans to merge Codex into ChatGPT in the coming weeks. Sottiaux adds that OpenAI generally prefers doing a series of small releases so it can get feedback as it goes, in part because the AI space moves so fast that “you can’t really afford to do a big splash and be wrong.”
Hundreds of millions of people in China and other countries have used super apps to do almost everything online for years. OpenAI is proposing a different vision, in part because it doesn’t really have any other choice.
WeChat and Alipay became ubiquitous by building the essential financial and information infrastructure that modern China now runs on. Countries like the US, on the other hand, already have Gmail and Instagram accounts, credit cards, and Venmo. As a result, OpenAI’s super app will likely have to find ways to plug into those preexisting systems.
OpenAI is making moves in this direction. Earlier this week, it announced an expanded partnership with Visa for agentic payments, and it previously built services that connect ChatGPT and Codex to your email inbox, Slack, and calendar.
OpenAI is ultimately betting that it can create a universal consumer interface so powerful and helpful that people no longer need to think about or interact with the websites, apps, and APIs underneath it. But that could leave it vulnerable to competitors who control the services and infrastructure it relies on. Sottiaux, though, is convinced that we’re headed toward a future where everyone has one agent that helps them navigate their entire life.
“OpenAI is known to take big, bold bets ahead of others, and this is us doing it again,” he says.
This is an edition of Maxwell Zeff’s Model Behavior newsletter. Read previous newsletters here.
此内容由惯性聚合(RSS阅读器)自动聚合整理，仅供阅读参考。 原文来自 — 版权归原作者所有。