OpenAI super app gets Thibault Sottiaux’s Codex engine plan, before an IPO pitch

OpenAI is betting its upcoming IPO story on one move that is both product and marketing: turning ChatGPT into a “super app,” an all-purpose AI interface that bundles OpenAI’s key products into a single agentic experience. The person driving this shift is Thibault Sottiaux, promoted from leading Codex to overseeing OpenAI’s core product platform for consumer, business, and developer customers. Speaking at VivaTech in Paris, Sottiaux said OpenAI wants to evolve ChatGPT from a conversational tool into a more personal assistant that remembers what users care about and gets better the more it learns from them.

In practical terms, his vision is that one unified agent can handle everything from quick questions to planning a complex trip and booking travel arrangements, then even spinning up a custom app to help a child learn trigonometry. Fortune reports the company plans to merge ChatGPT and Codex into “one unified agentic experience,” with Codex positioned as the engine that powers this broader interface. For an IPO audience, the subtext is clear: OpenAI needs to show investors it is not a single-product chatbot company, but a platform for ongoing tasks, personalization, and action.

This is not just a UI redesign. Over the past few months, OpenAI has reorganized product leadership to align the organization behind the bet. Product strategy is now under OpenAI president Greg Brockman, while Sottiaux leads core product and platform. Meanwhile, Nick Turley, previously head of ChatGPT, has moved to lead enterprise industries. The consolidation came with shutdowns and pullbacks that underline how serious this pivot is. OpenAI discontinued Sora, its AI video product, in April, and it halted plans for Stargate data center projects in the U.K. and Norway after previously considering expensive infrastructure commitments.

Those decisions matter because the “super app” pitch depends on execution discipline. Agentic products are expensive in every way: compute, tool access, safety work, and the product iteration cycle required to make autonomous actions actually work in the real world. If you want an agent that can browse, fill forms, take actions on a user’s behalf, or do anything beyond answering questions, you need models that can follow instructions reliably, interpret messy intent, and operate with appropriate permissions. That is exactly where Sottiaux draws the contrast.

OpenAI has tried agentic offerings before, and not all of them reached mainstream traction. Operator, OpenAI’s earlier attempt at an agentic product that could access the internet and take actions on the web for users, like browsing sites, filling forms, and completing tasks autonomously, largely failed to take off. Sottiaux explained that Operator was “early” and “didn’t quite work” because “the models were not” ready. The key difference now, he argued, is that the models can handle a larger set of tools, parse more genuinely ambiguous instructions, and operate with more permissions because of advances in ensuring the models accurately follow user intentions. He also pointed out that Operator often had to operate inside a constrained “sandbox” to reduce mistakes when given wider latitude for transactions.

The super app framing also connects to a broader global product lesson. Sottiaux said super apps are ubiquitous in China and much of Southeast Asia, where platforms like WeChat and Alipay bundle messaging, payments, food delivery, ride-hailing, and commerce. The user rarely needs to leave the app to get through daily life. In the West, that’s harder. Users in the U.S. and Europe are used to switching between specialized apps, and Western tech companies have largely failed to replicate the same pattern. OpenAI’s vision, according to Sottiaux, is not to force a single app to replace everything, but to build what he called a “personal AGI,” a unified interface that gets more capable and more personalized over time.

And yes, enterprise is the pressure test. OpenAI is trying to sharpen its case against rival AI lab Anthropic, especially as the enterprise market narrative has shifted slightly in Anthropic’s favor over the past year. Ramp’s May 2026 AI Index, which tracks actual corporate spending across thousands of companies, showed Anthropic at 34.4% of U.S. business adoption versus OpenAI’s 32.3%. The reversal, Fortune reports, was driven largely by Claude Code, Anthropic’s autonomous coding tool, while ChatGPT has been seen as more consumer-focused, a perception OpenAI is now working to change.

Still, an all-in-one agent is a tricky sell to large enterprises. Sottiaux said OpenAI does not see a hard divide between consumer and enterprise use cases for the new product. But there is an open question whether enterprise customers would rather get tailored, vertical solutions built for specific industries and workflows instead of a horizontal platform promising to do everything. Large enterprise buyers have typically been suspicious of horizontal platforms and often prefer integrations that slot into existing systems. Sottiaux’s answer is that the underlying technology is not fundamentally different; what changes is the ability to connect the right context, connect the right tools, and take actions that minimize risk to deliver value.

Strategically, this places OpenAI into a crowded “super app” conversation with real competitors. Microsoft is working on a one-stop shop that would connect GitHub Copilot, its Copilot chat function, the Cowork productivity tool, and a new agentic workflow capability into a single app, with plans to launch by the end of summer, Fortune previously reported. Elon Musk has also long held ambitions to turn X into a super app across communication, media, and commerce, citing the Chinese model as a template, according to several public statements.

For decision-makers, the stakes are straightforward and brutal. If OpenAI can make an agentic “super app” credible for both consumers and enterprises, it gains an IPO narrative that is larger than “chat.” If it cannot, it risks being valued like a single-interface company while Anthropic continues to pick off enterprise use cases with autonomous coding tools. In a wave where SpaceX, OpenAI, and Anthropic are all targeting combined valuations that add up to more than $3.6 trillion, the market will demand not just demos, but proof that agentic capability scales into everyday workflows. The question Sottiaux is trying to answer is whether OpenAI can turn ChatGPT into a general-purpose, personalized system that actually does work, not just talk, before the world decides what its public-market price tag should be.