OpenAI is doing what every tech giant eventually does when growth plateaus: it's pivoting to a platform play. The company is consolidating its scattered product lineup into a single, all-encompassing "super app"—a ChatGPT that doesn't just answer questions but supposedly manages your entire life. And the engineer now holding the controls is Thibault Sottiaux, freshly appointed as OpenAI's head of core products.

If you haven't heard the name yet, you will. Sottiaux now oversees both ChatGPT and Codex, and more critically, the project to fuse them into something OpenAI is calling—without a hint of understatement—the world's best personal agent. He reports directly to Greg Brockman, who stepped back into product oversight while CEO of AGI deployment Fidji Simo is on medical leave. In other words: Sottiaux is operating near the top of the food chain at one of the most heavily scrutinized AI companies on the planet.

No pressure.

Who Is This Guy, Actually?

Sottiaux isn't a product manager who stumbled into AI. He has actual engineering credibility, which is refreshingly rare in an industry that loves to promote people whose primary skill is narrating slide decks. He grew up in Belgium, studied applied mathematics, then landed at Google's London office in 2015 working on Maps. From there he moved to Google DeepMind, where he built the infrastructure scaffolding that researchers used to develop things like AlphaGo—the system that famously defeated a world Go champion in 2016 and made a lot of people suddenly nervous about their career choices.

When ChatGPT dropped in late 2022, Sottiaux apparently felt the ground shift in a way that DeepMind hadn't yet acted on. "This is something we had been sitting on at DeepMind for almost two years, and we were just not doing it," he said. He packed up, moved to San Francisco, and joined OpenAI in 2024.

His first assignment was familiar territory—building internal tooling for OpenAI's researchers, essentially the same role he had at DeepMind. But within months he pivoted to something with a much larger blast radius: Codex. The AI coding tool grew fast and made Sottiaux something of a cult figure among developers. He personally responded to bug reports on X and would occasionally intervene to reset weekly token limits for engineers who'd burned through their quota. That kind of hands-on responsiveness is either genuinely admirable or a sign that the support infrastructure was underpowered. Probably both.

The Super App Vision (And Why It's Harder Than It Sounds)

Here's the pitch: ChatGPT stops being a chat interface and becomes a proactive personal agent that deeply understands what you care about and delivers the right information at exactly the right moment. Sottiaux's phrase for it is "delightfully proactive." OpenAI wants you to think less WeChat clone and more omniscient digital chief of staff.

In practice, what that probably looks like is a highly capable assistant with persistent memory and broad integration hooks. Book a dinner reservation, yes—but also remember you had a bad reaction to shellfish last time, flag the expense report deadline before you miss it, autonomously run web lookups and API calls without surfacing the plumbing to the user. The user just speaks in natural language and things happen. That's the vision.

Under the hood, Sottiaux says Codex is doing most of the heavy lifting. The agent layer is essentially Codex repurposed as a general-purpose executor—it writes code, calls APIs, and browses the web as needed to complete tasks, then pipes results back through ChatGPT's familiar interface. From a systems perspective, that's actually a reasonable architecture. Codex already has strong task decomposition and tool-use capabilities, and it's apparently finding traction with non-technical users faster than anyone expected.

The consolidation strategy behind this is worth noting. To fund the super app push, OpenAI has shut down several standalone products—including the Sora video app and a dedicated platform for scientific research. The executives running those teams have mostly departed. This is a classic platform consolidation move: kill the periphery, concentrate resources on the core bet. It can work brilliantly or it can mean you've just thrown away the optionality that kept you interesting. History has examples of both outcomes.

The Part Where I Put On My Skeptic Hat

Let's talk about what the press releases conveniently omit.

Nearly a billion weekly active users is a staggering number to be responsible for. Sottiaux's entire career has been building tools for researchers and developers—people with high tolerance for rough edges, who file detailed bug reports and understand that "beta" means something. Consumer AI products are a different beast entirely. The failure modes are more embarrassing, the support burden is astronomical, and the margin for a bad experience is essentially zero before someone screenshots it and posts it to social media.

Then there's the "proactive" problem. An AI that surfaces the right information at the right time sounds elegant until you think about what "right" actually requires. It requires persistent memory that users will trust with sensitive personal data. It requires inference systems that can act autonomously without hallucinating a dinner reservation at a restaurant that closed two years ago. It requires latency low enough that the "agent completing your task" experience doesn't feel like watching a progress bar. These are hard, unsolved engineering challenges—not marketing checkboxes.

The super app concept also assumes OpenAI can out-integrate everyone else. Google has Maps, Search, Gmail, Calendar, and Workspace baked into its ecosystem. Apple has the entire iPhone. OpenAI has ChatGPT and a developer API. Building an agent that genuinely manages your life requires deep hooks into the platforms where your life actually lives. That's a partnership and integration problem as much as a model capability problem, and it's one where OpenAI starts significantly behind.

And then there's the IPO context. OpenAI has confidentially filed to go public. A super app narrative is exactly the kind of growth story that investors want to hear heading into that process. That doesn't mean it's not real—but it does mean you should calibrate accordingly when evaluating the ambition of the claims versus the current state of the product.

What Actually Matters Here

Here's what I find genuinely interesting, stripped of the hype: OpenAI is making a coherent architectural bet that a code-execution layer (Codex) is the right substrate for a general-purpose agent. That's not an obvious choice—lots of agent frameworks treat code generation as one tool among many rather than the core primitive. If that bet is right, Codex's strong developer adoption gives OpenAI a meaningful head start on reliable task execution, which is the part of agentic AI that most systems currently botch.

Sottiaux's technical background also matters. Engineers building for other engineers tend to over-index on capability and under-index on experience. The transition to consumer product thinking is real and it's genuinely hard. His self-described reaction—"incredibly exciting and mildly terrifying"—at least suggests he's aware of the gap he's crossing.

The core team is reportedly small, around 40 people as of a couple of months ago. For a consumer product targeting a billion users, that's either a sign of lean engineering discipline or a sign that the ambition dramatically outpaces the current headcount. Probably worth watching which direction that number moves as the IPO timeline gets closer.

Super apps have worked before, mostly in contexts where a single platform had regulatory and ecosystem advantages that made lock-in almost inevitable. OpenAI doesn't have those structural advantages in Western markets. What it does have is a model that a lot of people genuinely find useful and a brand that still carries serious weight. Whether that's enough to make "delightfully proactive" a real product description rather than a roadmap fantasy—that's the question Sottiaux now has to answer.