Keep pulling the thread on Cat Wu.
At Anthropic, the development timelines for many product features have been reduced from six months to as short as one month or even one day.
Anthropic has an internal model named Mythos that is still in preview due to its high power and capabilities.
A key factor in Anthropic's success is its unifying mission of bringing safe AGI to humanity, which helps in hiring and making fast, cross-organizational decisions.
When a new, more capable model is released at Anthropic, a significant part of the product development process involves removing features that were previously added as "crutches" for the less capable older models.
Anthropic's internal engineering team now relies on its AI-powered code review tool, built with the latest models, to pass checks before merging pull requests.
Anthropic's vision for its agentic products is to scale from single-task success to managing hundreds of remote tasks simultaneously, with a focus on verification and self-improvement from user feedback.
The major shift in AI products, exemplified by Claude Code, is the move from chat-based interaction to action-based agents that can perform tasks on a user's behalf.
Cat Wu believes that as code becomes cheaper to write due to AI, the most valuable skill is deciding what code to write, which she defines as "product taste."
The Claude Code team at Anthropic ships almost all of its new features under a "Research Preview" label to reduce the commitment and get user feedback quickly.
The recent leak of Claude Code's source code was the result of human error during a process where a person was using Claude to write a pull request.
Anthropic decided to restrict third-party tools like OpenClaw from using Claude subscriptions because the service was not designed for the different usage patterns of third-party products.
Anthropic's product management organization is divided into several teams: Research, Claude Developer Platform, Claude Code, Enterprise, and Growth.