OpenAI just raised $40 billion on the promise of artificial general intelligence. It is now dismantling the teams that built its most science-fiction products to focus entirely on writing code.
On Friday, OpenAI confirmed that Kevin Weil, the former Instagram VP who led the company's science application team, is departing as that team gets folded into Codex, OpenAI's coding agent product. Bill Peebles, who headed the Sora video generation team, also announced his exit—hours after OpenAI quietly killed Sora itself. The departures are not incidental. They are the human cost of a strategic about-face: OpenAI is abandoning consumer-facing AI to stake its future on a single product category.
The conflict is stark. OpenAI built its cultural caché and $40 billion valuation on the premise that transformative AI would arrive for everyone—the researcher, the artist, the student. Sora was the embodiment of that vision: a model that could generate minute-long videos from text prompts, the kind of product that made AI feel inevitable. Now it's gone. The science team is gone. The vision is gone. What remains is Codex—a coding tool for developers and enterprises.
This is not restructuring. It is a bet that agents are the only AI product with a defensible business model. The reasoning is financial, not philosophical. Enterprise customers pay subscription fees. Developers build workflows that lock them into platforms. Agents that complete complex tasks generate recurring revenue in ways that consumer products—freemium chatbots, image generators—simply don't. Every time OpenAI shipped a consumer feature, it competed with its own API customers and watched usage spike without corresponding revenue growth.
The science team represented the opposite model: speculative investments with unclear monetization paths. Sora launched and immediately faced questions about copyright liability, content moderation costs, and whether anyone would pay for video generation when free alternatives existed. Sam Altman framed it as "focus on mainline products over side quests." That's corporate-speak for admitting the side quests were consuming resources without producing returns.
Weil's departure is particularly telling. A former Facebook executive who successfully monetized Instagram's consumer base through advertising and shopping features, he came to OpenAI to apply that playbook to science applications. If he couldn't find a path to monetization at AI's most capitalized company, that's not a personal failure—it's evidence that consumer AI monetization remains unsolved at scale.
Developers and enterprise buyers win in this scenario. OpenAI will pour science team resources into Codex improvements, faster execution, better context windows, more reliable code generation. The company that promised AGI for everyone is now promising better autocomplete for engineers. The question is whether that narrow focus is wisdom or retreat—and whether a $40 billion valuation can survive on coding agents alone.
Sam Altman posted on X that the company would "keep building things that matter." Apparently, only one thing does.