Cat Wu, who led product on Claude Code, posted on X on November 3, 2026 that “Friday was my last day at Anthropic.” She is the third person from the Claude Code team to announce an exit in under two weeks, after engineer Thariq Shihipar and Anthony Morris, who built Claude Code on desktop, both posted on October 24.
All three said it in their own words, and none of them said where they are going. Here is what each wrote, the one line in Wu’s post that says the most about where coding agents went, and what Anthropic has not said.
Cat Wu posted on November 3, 2026 that Friday was her last day at Anthropic, where she worked on Claude Code from its internal-tool days.
Engineer Thariq Shihipar and Anthony Morris, who built Claude Code on desktop, both announced their exits on October 24, 2026.
Wu writes that at launch most people approved every permission prompt by hand; today the default is trusting Claude to auto-approve.
None of the three says why they left; Shihipar plans a break and Morris says he will share his next project soon.
We found no Anthropic statement on Wu’s successor or on any Claude Code change as of 09:20 UTC on November 4.
§ 01What each of them wrote
Wu’s post went up at 21:33 UTC on November 3 and had passed 1.6 million views on X by early November 4. She tells how she joined: “Two years ago, when Claude Code was a small internal tool, I asked it to build an RL env.” She “never manually wrote a line of code”, and “It worked, I was hooked, and I joined the team.” She thanks “the incredible team behind Claude Code, Cowork, and Tag.”
Shihipar posted at 04:04 UTC on October 24: “I’ve left Anthropic after more than a year of working on Claude Code and seeing it scale larger than I ever imagined.” His plan: “For me, it is time for a bit of a break before diving into something new.” He closes by pointing readers to colleagues still there, ending with Boris Cherny, “who started it all”.
Morris posted at 12:33 UTC the same day: “Yesterday was my last day at Anthropic.” He writes “I created Claude Code on desktop because I believe everyone can be a builder”, calls leaving “the hardest thing I’ve done in my career”, and says “I’ll share what’s next soon!”
| Person | Role, in their own post | Posted (UTC) | What’s next, in their words |
|---|---|---|---|
| Thariq Shihipar | Worked on Claude Code for more than a year | October 24, 2026, 04:04 | A break, then something new |
| Anthony Morris | Created Claude Code on desktop | October 24, 2026, 12:33 | Will share soon |
| Cat Wu | Part of the team from its internal-tool days; product lead per India Today | November 3, 2026, 21:33 | Not stated |
India Today, which reported the three exits together on November 4, describes Wu as product lead, Shihipar as an engineer on the coding assistant and Morris as the builder of its desktop version, and notes the timing ahead of Anthropic’s planned IPO. The departures come as Anthropic works through its public offering, which we track in our Anthropic IPO record.
§ 02The line about approvals
The sentence in Wu’s post that reaches past Anthropic is about permission prompts. “When we first launched, most people approved every permission prompt by hand and mostly used Claude to answer questions and write tests.” Then: “Today, the default is trusting Claude’s judgment to auto-approve. Claude does large migrations and builds new products.”
That is a product lead describing, in two sentences, the change in who says yes. Claude Code sessions now start in auto mode on Pro, Max and Team plans, with a safety classifier deciding which actions still stop for the user; we explained what it checks in our Claude Code auto mode guide. The same week, a Codex user reported that an agent asked to reboot one server went on to change its mounts and containers, which is the other side of the same trade; our Codex incident record reads that report.
| At launch, per Wu | Today, per Wu | |
|---|---|---|
| Permission prompts | Most people approved every one by hand | Default is trusting Claude’s judgment to auto-approve |
| Typical work | Answering questions and writing tests | Large migrations and building new products |
§ 03What Anthropic has not said
- Who leads product now. We found no Anthropic statement naming Wu’s successor as of 09:20 UTC on November 4.
- Why they left. None of the three gives a reason beyond Shihipar’s break and Morris’s next project.
- Whether anything changes for users. No announcement ties the exits to a Claude Code roadmap change. Morris points desktop feedback to two colleagues still at Anthropic.
Three exits from one product team in eleven days is a fact. Whether it is a pattern is something Anthropic has not addressed, and the three posts do not suggest a shared cause: each reads as a personal goodbye.
§ 04Who says yes, on an AI employee
Wu’s line about auto-approve is the question every team that hands work to agents ends up answering. On CellCog, the owner sets one approval level for an AI employee; anything the agent marks above that level waits for the owner’s yes, and everything else runs on its own. You can start strict and loosen it as the employee earns trust, which is the path Wu describes Claude Code’s users taking over two years.
§ 05Sources
- Cat Wu on X, November 3, 2026, 21:33 UTC.
- Thariq Shihipar on X, October 24, 2026, 04:04 UTC.
- Anthony Morris on X, October 24, 2026, 12:33 UTC.
- India Today, Anthropic loses 3 Claude Code team members, just weeks before its mega IPO, November 4, 2026.
Q1When did Cat Wu leave Anthropic?
Her last day was Friday, October 30, 2026, per her X post of November 3, 2026, which opens with Friday was my last day at Anthropic.
Q2Who else left the Claude Code team?
Thariq Shihipar, who worked on Claude Code for more than a year, and Anthony Morris, who created Claude Code on desktop, both announced their exits on October 24, 2026.
Q3Who leads Claude Code product now?
Anthropic has not said as of 09:20 UTC on November 4, 2026. Shihipar’s post credits Boris Cherny as the person who started Claude Code.
Q4Does this change Claude Code for users?
No change has been announced. Morris points desktop feedback to two colleagues who remain at Anthropic.
Q5What did Cat Wu say about permission prompts?
That most early users approved every prompt by hand and mostly used Claude for questions and tests, while today the default is trusting Claude to auto-approve and it does large migrations.