OpenAI loses three more executives in a week as it reorganizes ahead of an IPO

The departure of three prominent executives from OpenAI on April 17, 2026 looks less like routine staff rotation and more like the continuation of a deep internal restructuring. The startup, which until recently could afford to run scientific, consumer and experimental efforts in parallel, is now visibly trimming side initiatives and pulling the organization into a simpler structure. Against that backdrop, science VP Kevin Weil, Sora head Bill Peebles and Srinivas Narayanan, CTO of enterprise applications, are leaving the company.

Weil's exit is the most symbolic. He joined OpenAI in June 2024, first as chief product officer, then took over OpenAI for Science. That team was meant to show how OpenAI's models can speed up scientific research, especially in biology and medicine. Weil himself wrote that he sees accelerating science as one of AI's greatest strengths. But the company has now decided to decentralize that work: instead of a separate unit, the research will be spread across other product, research and infrastructure teams. In other words, the science vertical ceases to exist as an independent center of power.

Prism is being wound down along with it. It is a web app for scientists that OpenAI launched only in January 2026. According to Western media, the project's small team is being moved under Codex, and some of Prism's features are planned to be built into the desktop coding app. This illustrates OpenAI's new priority well: the company does not want to maintain standalone products with narrow audiences if the same technology can be built into a larger platform.

Notably, just before Weil's departure OpenAI introduced GPT-Rosalind, a series of models for life sciences and drug discovery. So science as a theme is not disappearing, but its form is changing: less standalone branding, more integration into the main product lines.

Bill Peebles's departure also reads in this context. He was responsible for Sora, one of OpenAI's most visible and high-profile video generation projects. But back in March the company effectively halted Sora's development as a standalone product. According to US press reports, the service burned about $1 million a day on compute alone, and the return from this effort fell short of expectations. In parallel, as previously reported, OpenAI cut other so-called side quests: ambitious but poorly monetizable experiments. In his farewell, Peebles stressed that such research needs space outside the main product roadmap. In business terms, that means one simple thing: OpenAI now has less tolerance for expensive experiments without quick, clear results.

The third departure, Srinivas Narayanan, who oversaw enterprise applications, matters especially because it comes just as OpenAI is betting on the enterprise market, development and platform products. According to US media reports, he told the company internally that he is leaving to spend more time with his family.

A broader management shift is also under way. Earlier, Fidji Simo, who was responsible for AGI deployment and launched a series of organizational reforms, took medical leave. Chief marketing officer Kate Rouch has also stepped away from work for health reasons. Chief operating officer Brad Lightcap moved to a role tied to special projects, and co-founder and president Greg Brockman is temporarily overseeing the product side. Even Sam Altman recently acknowledged that OpenAI can no longer operate as a chaotic startup and must become a more predictable large platform.

What this means: OpenAI is rapidly turning from a lab with many parallel bets into a more tightly managed product company. Priorities are getting clearer: enterprise, developer tools, large platforms and areas that can be scaled quickly and defended in front of investors. But this turn has a price. When a company closes side tracks and disbands autonomous teams, it also reduces internal diversity and risks losing the people who created the boldest experiments.

So the current string of departures matters not only as news about individuals. It shows what kind of OpenAI the company wants to be in its next phase: more disciplined, more commercial and noticeably less tolerant of expensive detours from the main course.