Cat Wu announced on X on November 3, 2026 that she had left Anthropic, where she led product for Claude Code. The departure closes a two-year period in which the coding tool moved from an internal experiment toward more autonomous software development. Wu thanked the teams behind Claude Code, Cowork and Tag, calling the products the result of thousands of ideas from colleagues and user feedback. She described building Claude Code into an industry-defining product as the journey of a lifetime. For companies, the event matters because the tool now defaults to trusting the model to act with fewer interruptions.
From internal experiment to auto mode
Wu joined Anthropic in August 2024 as a product manager on the Research PM team, which linked researchers with real-world customers to improve models. Before that she worked as a product engineer at Scale AI and Dagster, then as a venture capitalist, while continuing to write code that tracked company announcements and surfaced open-source projects gaining momentum. When Claude Code became available internally in fall 2024, she used it to build Streamlit applications for analyzing large-scale user feedback, run evaluations for potential benchmarks and create reinforcement-learning environments. Those projects took hundreds of hours of prompting with Sonnet 3.5 (new), without writing code by hand. She also described testing models on the same task of adding a table tool to Excalidraw across successive releases.
The Excalidraw test traced model progress from failure to reliable demonstration. With Sonnet 3.5 (new) in October 2024, Claude still failed the task, according to Wu's March 19, 2026 account. Opus 4 occasionally succeeded in June 2025, and Opus 4.6 later completed such feature requests reliably enough for live demonstrations before thousands of professional developers. Anthropic formalized that capability shift in its auto-mode announcement of August 7, 2026, updated October 22, 2026. New sessions on Pro, Max and Team plans were set to run in auto mode from August 14, 2026, with the default extending during the following week to new sessions in most Claude apps on those plans. Instead of approving each action, users let safe tool calls proceed automatically.
The mechanism centers on a classifier that filters tool calls by risk. Auto mode routes calls through the classifier to block irreversible or destructive actions and actions aimed outside the user's environment, while allowing safe actions to proceed. When an action is blocked, Claude can seek a safer route or ask for approval. Three consecutive blocks, or 20 blocks across a session, return Claude Code to manual approvals. Anthropic cautions that classification does not eliminate risk and recommends direct review of high-stakes changes to production infrastructure. The practical change for users is a redistribution of oversight, with routine actions moving faster and production changes still requiring human control.
What autonomous coding means for business
For companies that build software with AI assistance, the shift reduces friction in everyday development work. Answering questions and writing tests under manually approved permission prompts has given way to auto-approved migrations and new product builds as the default described by Wu. Small teams can move through routine implementation with fewer pauses for confirmation, which shortens feedback loops on prototypes and internal tools. Larger organizations can apply the same autonomy to large migrations across codebases, provided ownership of review is clearly assigned. The difference lies in coordination: a small firm feels the gain in speed, while a large firm feels it in throughput across many developers.
The same autonomy introduces conditions that buyers should verify before widening access. The classifier is designed to stop destructive or out-of-environment actions, but Anthropic states that it does not remove risk entirely. Teams should keep manual review for production infrastructure, data deletion, access changes and customer-facing releases. Wu's account also shows a changed way of working, where designers shipped code, engineers made product decisions and managers built prototypes, with short experiments replacing longer documentation-first cycles. Features such as Claude Code on Desktop, the AskUserQuestion tool and todo lists emerged from those experiments. The departure itself does not signal a change in product direction, so customers should judge the roadmap by shipped defaults and documented safeguards.
A concrete marker to follow is how broadly the auto-mode default holds across Pro, Max and Team plans after the October rollout week. If new sessions stay in auto mode and manual fallbacks remain rare, autonomous execution will have become the normal operating pattern. A second signal is who takes over product leadership for Claude Code, Cowork and Tag. That appointment will show how Anthropic intends to balance speed of execution with supervision in the next phase.
