Cat Wu ran product for Claude Code and Cowork and will now lead product management for OpenAI's Core Products group. Anthropic is weeks from its IPO. OpenAI also agreed Wednesday to buy Pierre, a startup whose tools already run in Codex, and AMD's Sharon Zhou said she had joined Anthropic.

Cat Wu, who led product for Anthropic's Claude Code, is going to OpenAI. She posted the news on X at about 4:25 p.m. Pacific on Wednesday, one day after she wrote that Friday had been her last day at Anthropic.

She'll run product management for OpenAI's Core Products group and work with Thibault Sottiaux, the OpenAI product head she tagged in her post. According to RuntimeWire, the group covers ChatGPT, Codex, the API and OpenAI's agent infrastructure. Wu wrote that she wants to build agents "that have your back and your team's" and that keep people in control.

OpenAI president Greg Brockman posted a welcome. Wu's post had about 2,000 likes within half an hour, and X posts mentioning her jumped from about 25 an hour to more than 570.

Wu was one of the two faces of Claude Code

Wu joined Anthropic in August 2024. Business Insider describes her and Boris Cherny, who created Claude Code, as the product's two main leaders and promoters. She also ran Cowork, which uses the same models for work outside of coding.

She has left once before. The Information reported that Wu and Cherny went to Cursor in the summer of 2025 and returned to Anthropic a few weeks later.

In Tuesday's goodbye post, Wu wrote that she joined the team two years ago after asking what was then a small internal tool to build a reinforcement learning environment, without writing any of the code herself. She also wrote that when Claude Code launched, most users approved every permission prompt by hand. Today the default is letting Claude approve its own actions.

Wu is the third Claude Code team member to announce an exit in under two weeks. Engineer Thariq Shihipar and Anthony Morris, who built Claude Code's desktop app, both posted on Oct. 24. Shihipar said he'd take a break first, and Morris said he'd share his plans soon. Anthropic hasn't said publicly who now leads product for Claude Code.

The IPO pitch leans on Claude Code

Anthropic filed its S-1 on Oct. 26 and is expected to list on Nasdaq under the ticker ANT before the end of November, as we reported last week. Business Insider credits Claude Code with giving Anthropic a breakout product. Developers ran up big token bills on it, Anthropic's revenue grew, and the company became a stronger competitor to OpenAI.

Anthropic's bankers will spend the next several weeks marketing the company with Claude Code's product lead now working at OpenAI. Goldman Sachs, one of the lead underwriters, restricted Claude Code to its own engineers in October.

OpenAI spent the same day buying more of the coding stack

Hours before Wu's post, OpenAI said it would acquire Pierre Computer Company, a three-year-old startup that builds Git storage for AI agents. Pierre's open-source Trees and Diffs libraries already run in the Codex app, where they handle file browsing and change review. OpenAI didn't disclose a price, and the deal still needs regulatory approval.

Pierre was founded by Jacob Thornton, a co-creator of Bootstrap, and Ian Ownbey, Twitter's first intern. Both said Wednesday they're joining OpenAI. The company raised a $22 million Series A from CRV and 01A a year ago.

OpenAI agreed earlier this year to buy Astral, which makes the Python tools uv and Ruff, and in June agreed to buy Ona, which runs cloud sandboxes for agents. In its Pierre announcement, OpenAI said it wants Codex to be where teams build software together, which lines up with the team agents Wu described in her post.

Anthropic also hired from AMD

Anthropic made a hire of its own the same day. Sharon Zhou, a vice president of engineering and AI at AMD who worked directly with CEO Lisa Su, said on X that she had joined Anthropic to return to frontier research. Zhou founded the enterprise AI startup Lamini and moved to AMD with several Lamini employees in 2025.

In July, when AMD expanded its partnership with Anthropic, Zhou posted that Anthropic would run about 2 gigawatts of AMD GPUs. Her work focuses on getting models to write and tune GPU kernels, the low-level code that determines how efficiently a chip runs. It's related to the chip work Anthropic pitched to IPO investors this week.

Our take

At State of AI, we don't think one departure is a reason for companies to rethink Claude Code. For companies trying Codex, Wu's hire says a lot about where OpenAI wants to take it.

Claude Code's quality depends mostly on Anthropic's models and on settings like permission defaults, and those don't change because a product lead leaves. Ask your Anthropic account team who owns Claude Code product now, and whether anything they promised you on the roadmap has a new owner.

OpenAI is putting code storage, review and sandboxes inside Codex so teams can do more of their software work there. If your engineers work in GitHub and you're testing Codex, ask how review and storage will work after Pierre is folded in, and what you'd need to move.

Wu's goodbye post described users moving from approving every action to letting the agent approve its own. Her OpenAI post talks about agents that keep you in control. Before any coding agent goes out widely at your company, find out which actions it takes without asking and who is allowed to change that setting.