OpenAI executive Kevin Weil exits as Prism folds into Codex

OpenAI is aggressively cutting its side quests, and the cleanup is claiming some of its most visible leaders.

On Friday, three OpenAI executives announced their departures in rapid succession. Former Chief Product Officer Kevin Weil, who most recently ran OpenAI for Science, confirmed his exit. Bill Peebles, who spearheaded the Sora video generation team, said his goodbyes on X. Internal enterprise applications CTO Srinivas Narayanan also shared that he is stepping down to spend time with family.

Alongside the exits, OpenAI is killing Prism, the dedicated web workspace for scientists it launched just three months ago in January. The roughly 10-person group behind Prism is being folded under Thibault Sottiaux’s Codex team, with its features destined for the Codex desktop app.

Folding science and video into Codex

The move marks a sharp pivot in OpenAI's product blueprint. Last month, the company pulled the plug on its standalone Sora video app, which TechCrunch reported was burning an estimated $1 million a day in compute. Now, scientific tools are facing the same consolidation.

Weil joined OpenAI in June 2024 as CPO before moving to lead science research last September. His tenure wasn't without drama. Last fall, Weil had to delete a post claiming GPT-5 had solved 10 unsolved Erdős math problems after the mathematician maintaining the database called out the error.

OpenAI isn't ditching scientific work entirely—it just rolled out its GPT-Rosalind model for life sciences researchers. But the dedicated consumer-facing wrapper is dead. A spokesperson confirmed that the goal is unifying product strategy, with the broader aim of turning Codex into an enterprise "everything app."

Pre-IPO tightening meets executive churn

These departures arrive amid widespread executive reshuffling. With AGI deployment chief Fidji Simo and CMO Kate Rouch both out on medical leaves, cofounder Greg Brockman has taken interim control of product. COO Brad Lightcap shifted to a special projects role, and CEO Sam Altman recently acknowledged that OpenAI has to operate as a predictable major platform, not a chaotic startup.

With Anthropic breathing down its neck in coding and an IPO planned for later this year, OpenAI cannot afford sprawling, costly experiments. Yet Peebles offered a poetic warning in his farewell, arguing that research labs thrive on cultivating entropy rather than letting everything "mode collapse to the most important thing."

Why it matters

OpenAI is officially done playing the freewheeling research lab. We think this consolidation makes cold, hard business sense for an upcoming IPO: bleeding $1 million a day on video demos or running boutique science frontends simply doesn't pass Wall Street muster when Anthropic is dominating lucrative enterprise code accounts.

For developers and power users, Codex absorbing Prism is actually good news. You won't have to bounce between isolated niche web portals; scientific modeling and analytical workflows will live inside a single, robust desktop powerhouse.

However, the strategy carries serious long-term risk. Breakthroughs like Sora only happened because OpenAI once allowed small, passionate teams to wander far off the mainline roadmap. By pruning away every high-variance research bet to juice enterprise software margins, OpenAI looks more like a standard Big Tech vendor every day—predictable, profitable, but potentially far less capable of catching the next big accidental paradigm shift.