OpenAI hires former product lead of Claude Code
Cat Wu joins OpenAI to lead product management of Core Products, after leaving Anthropic. Her arrival comes as agents capable of delegating tasks become a central battleground between the two rivals.
Cat Wu joined OpenAI on November 5, 2026, to take charge of product management for the Core Products, after leaving Anthropic three days earlier. This recruitment highlights the battle for AI products capable of acting on behalf of the user, but without taking control away from them. The challenge is simple to state, but harder to achieve: delegate more, while keeping control.
A new position that says a lot about the OpenAI-Anthropic rivalry
At OpenAI, Wu will work with Thibault Sottiaux, identified as head of core products, and with the company’s product teams. Her arrival is presented as a reinforcement to design agents that take on part of the work, while letting the user decide, verify, and correct. Greg Brockman, president and co-founder of OpenAI, welcomed her recruitment, while Sottiaux greeted her with a wink on X: “CatGPT.”
The move also matters on the competitive front. OpenAI and Anthropic are competing not only for models, but also for talent, use cases, and the way to sell AI as a concrete product. In this context, recruiting a manager who led Claude Code is not a mere hiring detail, it is a way to buy experience where it is hardest to replicate: in product trade-offs.
Claude Code, from an internal tool to a closely watched product
Before this departure, Wu explained that she had joined Anthropic two years ago, when Claude Code was still just a small internal tool. She recounts launching it after asking Claude to create a reinforcement learning environment, then guiding it for dozens of hours without writing any code herself. At the time, users often manually approved each authorization request. More recently, according to her account, the product switched to default auto-approval, with more ambitious uses such as massive migrations and the creation of new products.
This journey sums up well the climb for software agents. An agent is a system capable of performing tasks with a higher degree of autonomy than a simple chatbot, but its value depends on clear limits. Wu says she wants to build agents that “take work off users' plates while keeping them in control,” a phrase that sums up the dilemma of the entire sector: more autonomy, without turning AI into an unruly intern. The subject makes one smile for a moment, then quickly returns to the reality of businesses.
The real issue: making AI work longer, but under control
At OpenAI, this logic fits into a strategy already visible around Codex and the planned integration of Ona. The company stated on June 11, 2026, that more than 5 million people used Codex each week, a 400% increase compared to early 2026, and that the tool was no longer used only for software development. It is also employed for research, analysis, system building, and automation, with tasks now extending over several hours or even days.
The Ona project aims to provide persistent environments where agents maintain access to tools, systems, and context over time. Practically, this allows a task to run in the company’s cloud even when the laptop is closed. The user can then track progress, give direction, make decisions, and review results remotely, rather than being tied to a single session.
What Wu brings to OpenAI, beyond the CV
Her arrival comes with very concrete experience. In organizational deployments, decisive criteria include execution location, authorized access, identity framing, action logging, and validation workflow. Ona’s controlled execution model fits precisely this logic: the agent operates in the organization’s cloud environment, while OpenAI provides intelligence and orchestration. Here lies the continuity with Wu’s message on user control: delegation only makes sense if the framework is solid.
OpenAI has not yet said how this arrival will translate into its products, but the direction is clear. Assistants are no longer just tools that respond; they become systems that execute. The real question remains, more down-to-earth: who sets the safeguards, and at what point does the agent ask for human input rather than pressing the button alone?
Key points
- November 5, 2026: Cat Wu joins OpenAI.
- She left Anthropic three days earlier, according to her post.
- Codex exceeded 5 million weekly users in June 2026.
- OpenAI reported a 400% increase on Codex.
- Thibault Sottiaux oversees Core Products at OpenAI.
By the numbers
- 5 million — weekly Codex users worldwide, June 2026, OpenAI.
- 400% — Codex growth compared to early 2026, worldwide, June 2026, OpenAI.
- 2 years — time Wu spent on Claude Code before leaving, according to her account.
- 3 days — gap between her departure from Anthropic and arrival at OpenAI.