Anthropic’s Daily Iteration Model: 90s-Born Chinese-American Executive Catherine Wu on Token Costs B

While most companies still roll out new products on a quarterly basis, anthropic has compressed its release cycle to “daily iterations.” At the center of this shift is catherine “Cat” Wu, a product leader overseeing Claude Code and Cowork. In a recent in-depth interview, she revealed how the company accelerated development, reDeFined the PM role, and leveRAGed AI cost-effectively.

Key Interview Insights:

Accelerating the Release Cycle
Anthropic slashed feature development from 6 months to as little as 1 day. According to Wu, this speed comes less from using their own powerful models (like Mythos) and more from streamlining processes. The team rEMOved all bureaucratic bARRiers, enabling engineers and PMs to turn an idea into a shipped product within a week.

The Evolving Role of Product Managers
Wu, who interviews hundreds of PM candidates, noted that traditional long-term roadmap alignment is becoming obsolete. In the age of cheap code generation, the most valuable Skill is “product sense”—the ability to decide what to build, define a clear vision, and drastically shorten the distance from idea to user delivery. Strict roles are blurring; the best team members are those with a strong engineering and product mindset who also possess “first-principles thinking.”

Cowork and Inbox Zero
Wu demonstrated the power of Cowork, Anthropic’s tool for non-code ouTPUts. She described using it to prepare a complex presentation: by connecting it to Slack, Gmail, and Google Drive, Cowork autonomously gathered context, compiled release notes, and generated a professionally designed 20-slide deck overnight. She emphasized that AI tools are most effective when used daily to automate repetitive, disliked tasks until they reach 100% reliability—not merely 95%.

Token Economics and Security
Internally, the APPlied AI team logs the highest token usage outside of engineering. Wu confirmed that while per-capita Token Consumption spikes with every major model upgrade, the cost remains far lower than an engineer’s average salary. On the recent claude Code source code leak, she clarified it was a non-malicious human error during a routine update review. The employee remains with the company, and processes have been strengthened.

Mission-Driven Sacrifice
Wu stressed that the team’s guiding principle is bringing safe AGI to all humanity. In a hierarchical sense, Anthropic’s success comes before any single product’s success: “If Claude Code fails but Anthropic succeeds, I would be very happy.” This unified mission allows the team to make quick, self-sacrificing decisions, including deprioritizing features or restricting third-party access to prioritize core products.

Outlook: From Tasks to Multi-Agent Management
Looking ahead, Wu sees the evolution moving from making a single task succeed to managing hundreds of parallel Claude sessions. The next challenge is building interfaces that help humans monitor, verify, and continuously improve these agentic workflows.