Cat Wu's move from Anthropic to OpenAI shows the coding-agent war is now a fight for the people who build the products

Cat Wu, who led product for Claude Code at Anthropic, is joining OpenAI. Business Insider reports she will run product management for OpenAI's core products team. She announced the move on X, and OpenAI confirmed it. Anthropic declined to comment. According to an OpenAI representative, Wu will work with Nick Turley on enterprise products and Ashley Alexander on consumer products, and will report to Tibo Sottiaux.

The timing makes this more than a routine hire. Wu said she left Anthropic last week. Anthropic is set to go public in a couple of weeks and is pitching its financials and growth story to investors. Claude Code is a big part of that story. Business Insider describes it as a tool that went from an internal prototype to a "commercial juggernaut" that has boosted both Anthropic's revenue and the daily work of software engineers. Wu was also one of its most visible champions on social media. On X, she wrote that she had "no idea" whether leaving would let her sell her shares before remaining employees could, and that it "wasn't a factor for me."

This is not the first time she has left. In 2025, Wu and Claude Code creator Boris Cherny went to the AI coding startup Cursor and returned to Anthropic shortly after, according to The Information. The two then led Claude Code's fast growth in late 2025 and into 2026, helped by better code-writing models. Developers moved to the tool, and later to OpenAI's equivalent, Codex, as each improvement automated more of the programming process. Business Insider notes that Codex has also driven OpenAI's growth and is now being aimed more and more at non-technical work.

**Our thesis:** this hire shows where the competition now sits. The gap between the top models is narrow. On our Zendoric Quality index, Anthropic and OpenAI lead the frontier, with China close behind. When models are that close, the edge comes from product: how you package an agent, how you earn trust, how you decide what the human controls and what gets automated. That know-how lives in a small group of people, and it can be hired away in a way model weights cannot. OpenAI isn't buying code. It is buying a playbook that has already worked.

**Our reading:** we shouldn't overstate the damage to Anthropic. One executive does not make a product, and the outflow we can verify here is a single person, not a pattern. But the signal ahead of the IPO is real. Retaining key talent is the kind of risk Anthropic's own S-1, at 93 pages of risk factors, should help investors audit. More interesting is the language Wu used to announce her move: "agents that have your back and your team's, and keep you in control." That fits a trend we keep seeing across the industry. In business agents, the value is shifting from raw capability to supervision, permissions and human control. If OpenAI's main products take on that focus, the user benefits, whoever wins the talent battle.

The longer view points the same way. Competition for the people who know how to make agents useful and safe speeds up the whole sector. In the short term it means turnover, fragile retention stories and pressure on valuations. In the long term it means better tools, built faster, that take on the mechanical parts of work and leave more time for what Wu herself called "the people and work you love." That outcome is not guaranteed. It is the right goal to judge these companies against, and their products will show whether they deliver.