🧭 Cat Wu Departs Anthropic: Three Claude Code Leaders Gone in Fourteen Days

Cat Wu, Head of Product for Claude Code and Cowork, has left Anthropic — the third named Claude Code team member to depart in a fortnight. Wu announced on November 3 that October 30 was her last day, following the exits of engineers Thariq Shihipar and Anthony Morris, who both said goodbye on October 24. Wu joined Anthropic in mid-2024 and quickly became one of the two most visible faces of Claude Code alongside its creator Boris Cherny, regularly appearing at developer events and in launch materials. Anthropic has not issued a public statement on her exit, nor announced a replacement product lead.

What it means for teams building on Claude Code

• No product pivot signalled: Anthropic has not changed its Claude Code roadmap or pricing publicly. The engineering team's core contributors remain, and the product continues to ship releases at its established cadence (v2.1.317 landed yesterday).
• IPO proximity: The departures arrive as Anthropic completes its IPO process, which is a period of compensation re-evaluation and vesting cliffs. Industry observers note this is a common attrition pattern in pre-IPO companies, not necessarily a signal of product-directional disagreement.
• Cursor connection: Business Insider, which broke the story, notes that Boris Cherny returned to Cursor earlier this autumn. Wu's next move has not been disclosed.

🧭 Message Threads in Public Beta — Server-Side Conversation Caching Cuts TTFT by 35%

Anthropic has launched Message Threads in public beta on the Messages API — a server-side conversation caching mechanism that eliminates the need for clients to resend full conversation history on every agent turn. For conversations in the 500 K–1 M token range, median time to first token drops by approximately 35%, with Claude Code already defaulting to threads for most users. The feature is available on the native Claude Platform and all three cloud offerings: AWS, Microsoft Foundry, and Google Cloud (Anthropic managed).

How it works

• The first request sends the full conversation and starts the thread. Add the beta header message-threads-2026-08-12 and a thread parameter with any identifier you choose.
• Subsequent requests include only the new messages and the previous response ID (previous_response_id). The API retrieves the cached context server-side.
• Threads are held in encrypted memory, not written to disk. They expire 5 minutes after the last request; your client falls back to resending the full history when a thread expires.
• Threads are append-only — you cannot edit, delete, or reorder prior turns once a thread is started. Each response ID can be continued only once.

import anthropic

client = anthropic.Anthropic()
BETA   = "message-threads-2026-08-12"

# Turn 1 — seed the thread
r1 = client.beta.messages.create(
    model="claude-sonnet-4-6",
    max_tokens=1024,
    messages=[{"role": "user", "content": "Summarise our Q3 metrics."}],
    betas=[BETA],
    thread={"id": "analytics-session-001"},
)

# Turn 2 — send only the new message
r2 = client.beta.messages.create(
    model="claude-sonnet-4-6",
    max_tokens=1024,
    messages=[{"role": "user", "content": "Now project Q4 based on that."}],
    betas=[BETA],
    thread={"id": "analytics-session-001"},
    previous_response_id=r1.id,
)

Current limitations (beta)

• Not supported: Message Batches, token counting, compaction on demand, context editing tools, and organisations with Inference hooks enabled.
• Threads cannot be listed, inspected, or deleted via the API — they are ephemeral by design.
• The 5-minute expiry restarts on every response the API stores, not on every request — keep this in mind for slow human-in-the-loop pipelines.

🧭 Anthropic Residency Opens Applications — $350K–$500K for Senior Technical Leaders to Work on AI Safety

Anthropic has opened applications for its next Residency cohort, a paid fixed-term programme targeting professors in quantitative disciplines, technical founders, and staff-level engineers who want to pivot into AI safety research. The programme runs between six weeks and six months, with annualised compensation of $350,000–$500,000. No machine learning background is required — the stated goal is to bring world-class technical rigour from adjacent fields (physics, mathematics, software systems, biology) into AI safety work. Applications close 22 November 2026.

Who it is designed for

• Senior technical leaders with demonstrated ability to tackle difficult, open-ended problems — not junior researchers building their first ML project.
• People with domain depth outside ML: interpretability work benefits from neuroscience and statistics, alignment theory from philosophy and decision theory, evaluation design from psychometrics and human factors.
• Founders or engineering leaders who have shipped complex systems and want to apply that experience to safety tooling, red-teaming, or capability elicitation methodology.

Programme structure

• Dedicated self-study time, internal talks, and direct mentorship from Anthropic research staff.
• Residents work alongside full-time teams rather than in a separate cohort bubble — the aim is real research output, not a lecture series.
• Anthropic explicitly states the goal is for every resident to leave capable of pursuing AI safety research full-time, either at Anthropic or elsewhere.