Claude Code product lead Cat Wu leaves Anthropic after two years

Cat Wu, head of product for Anthropic’s Claude Code, announced on X on November 3, 2026, that she has left the company after two years. She gave no reason for leaving and named no next role, and Anthropic confirmed her exit, according to HyperAI’s summary of the news.

Wu also oversaw Cowork, Anthropic’s knowledge-work tool for people who are not software engineers. Her departure lands during what HyperAI calls a critical stretch for the company, as it moves toward a planned initial public offering later this month. For developers who rely on Claude Code daily, the immediate question is less about the person and more about continuity: who sets the direction for a tool that has grown into one of Anthropic’s most important products.

What Cat Wu said in her farewell post

In her post, Wu thanked the teams behind Claude Code, Cowork and Tag, describing the products as the sum of thousands of ideas from colleagues. She credited user feedback with shaping much of what the teams shipped, and HyperAI reports she said she was grateful to the Claude Code team and its users. Wu characterized helping build Claude Code into an industry-defining product as the journey of a lifetime, according to Unite.AI.

What the post does not contain is equally notable. Wu did not explain why she is leaving. Neither source reports a successor for her role, a new employer, or any change to Claude Code’s roadmap tied to the departure.

HyperAI describes Wu as one of Claude Code’s two top public leaders, alongside the tool’s original creator, Boris Cherny. The sources do not say whether Cherny’s role is changing.

From research product manager to Claude Code lead

Wu’s own account of her product-management work, published March 19, 2026, traces how she reached the role, according to Unite.AI. She joined Anthropic in August 2024 as a product manager on the Research PM team, which connected researchers with real-world customers to improve models. HyperAI adds that she joined after testing an early prototype of Claude Code.

Before Anthropic, she worked as a product engineer at Scale AI and Dagster, then as a venture capitalist. In that investing role she kept writing code, building tools that automated monitoring of company announcements and identified open-source projects gaining momentum.

Using Claude Code before most people could

When Claude Code became available internally in fall 2024, Wu used it to build Streamlit applications (a Python framework for quick data apps) that analyzed large-scale user feedback. She also used it to run evaluations assessing potential benchmarks and to create reinforcement-learning environments to better understand training. Wu reported that these projects took hundreds of hours of prompting with Sonnet 3.5 (new), without her writing any of the code by hand.

The product reached the public later: HyperAI reports that Claude Code was released in research preview in February 2025.

One test task, repeated across model generations

Wu tracked model progress with a single repeatable exercise: adding a table tool to Excalidraw, the open-source whiteboard app. Her account, as summarized by Unite.AI, runs as follows:

• October 2024, Sonnet 3.5 (new): Claude still failed the task.
• June 2025, Opus 4: the model occasionally succeeded.
• Later, Opus 4.6: the model completed such feature requests reliably enough for live demonstrations before thousands of professional developers.

This is a personal, informal test reported by Wu herself, not an independent benchmark. It is still a useful signal of how the product team judged readiness: by whether a real feature request in a real codebase could be completed end to end.

How the Claude Code team worked under Wu

Wu’s account connects model progress to changes in how the team itself operated. According to Unite.AI’s summary, designers shipped code, engineers made product decisions and product managers built prototypes and evaluations. Short, self-directed experiments and demos displaced longer documentation-first cycles.

Several features developers now use came out of that experimental approach, including Claude Code on Desktop, the AskUserQuestion tool and todo lists. The AskUserQuestion tool lets the agent stop and ask the user for input rather than guessing, while todo lists let it lay out and track multi-step work.

Growth followed. HyperAI reports that Claude Code quickly became one of the company’s most important offerings, driving developer adoption, increasing token usage among tech customers and strengthening Anthropic’s competitive position against OpenAI. It describes Claude Code as a widely used product that helped attract revenue and funding for additional compute capacity.

Cowork, the second product Wu oversaw, builds on the same model capabilities and targets broader knowledge work for non-engineers.

The auto mode shift she leaves behind

Wu’s farewell contrasted Claude Code’s early days, when it relied on manually approved permission prompts and was used for tasks such as answering questions and writing tests, with what she described as the current default: trusting Claude’s judgment to auto-approve, conduct large migrations and build new products.

The product mechanics behind that shift are documented in Anthropic’s auto mode announcement, published August 7, 2026, and updated October 22, 2026. Instead of asking users to approve each action, auto mode routes tool calls (the commands and file edits the agent wants to perform) through a classifier, a separate model that sorts each action as safe or risky.

Auto mode detail	As reported
Announcement published	August 7, 2026; updated October 22, 2026
Initial date set	August 14, 2026, for new sessions on Pro, Max and Team plans
October update	Default to roll out the following week for new sessions in most Claude apps on those plans
What gets blocked	Irreversible or destructive actions, and actions aimed outside the user’s environment
When an action is blocked	Claude can seek a safer route or ask for approval
Fallback to manual	Three consecutive blocks, or 20 across a session

Anthropic cautions that the classification systems do not eliminate risk, and it recommends that users review Claude’s actions themselves for high-stakes changes to production infrastructure. In practice, Unite.AI notes, oversight is redistributed: routine actions move forward with fewer interruptions, while high-stakes production changes still call for direct human review.

Why the timing draws attention

Wu’s exit is not an isolated move. HyperAI reports that Claude Code contributor Thariq Shihipar also left Anthropic last month. Two departures from the same high-profile product team in quick succession will draw scrutiny, though neither source suggests the two are connected, and neither departing person has given a public reason.

The bigger backdrop is Anthropic’s planned public offering, which HyperAI says is expected later this month. Unicorn API covered the filing itself in Anthropic IPO filing shows $895B spending plan, warns of self-speeding AI. Senior product departures ahead of a listing are routinely watched by investors and customers alike, because they can signal shifts in strategy or simply the end of a natural tenure. The sources give no basis to say which applies here.

Claude Code has also faced questions about packaging and policy this year. The New Stack ran a report headlined “Anthropic is thinking about removing Claude Code from its cheapest plan”. Unicorn API has separately covered Anthropic’s confirmation of a game-clone rule in Claude Code and user reports of wider refusals, in Anthropic confirms Claude Code game-clone rule; users report wider refusals. None of the sources link Wu’s departure to any of these issues.

What it means for developers using Claude Code

For teams that build on Claude Code, nothing in the coverage points to an immediate product change. Auto mode, Claude Code on Desktop, the AskUserQuestion tool and todo lists all shipped under Wu’s tenure, and no source reports any of them being rolled back or altered because of her exit.

The practical considerations are about direction rather than features:

• Product philosophy. Wu was a public voice for a specific approach: fast experiments, demos over long specs, and steadily more autonomy for the agent. Whoever takes over product decisions may keep or adjust that balance.
• Autonomy defaults. The move toward auto-approval is the most consequential change for day-to-day use. Anthropic’s own guidance still asks users to review high-stakes production changes themselves, and that advice holds regardless of who leads the product.
• Cowork’s roadmap. Because Wu also oversaw Cowork, teams evaluating it for non-engineering staff may want to watch for any change in how Anthropic positions it.
• Feedback channels. Wu credited user feedback with shaping much of what shipped. Developers who have relied on public product leaders to raise issues may find those channels shifting.

The open questions are straightforward: who replaces Wu, whether Anthropic will name a single product lead for both Claude Code and Cowork, and whether the pace of autonomy-focused features continues. The sources answer none of them yet.

What to watch next

• A successor announcement. Neither source names who will lead product for Claude Code or Cowork.
• Wu’s next move. She named no next role. Given her background in product engineering and venture investing, a new company, investment role or rival lab are all possibilities, but nothing is reported.
• Further departures. Shihipar’s exit last month and Wu’s now make any further moves from the Claude Code team more newsworthy.
• The IPO. Anthropic’s planned offering later this month may bring new disclosures about leadership and the importance of Claude Code to the business.
• Auto mode rollout. The October update set the default to roll out to new sessions in most Claude apps on Pro, Max and Team plans. How that lands with users is the clearest near-term test of the direction Wu helped set.