How Anthropic’s product team moves faster than anyone else | Cat Wu (Head of Product, Claude Code)

Cat Wu emphasizes the challenge of being 'the right amount of AGI-pilled,' highlighting that while building for super AGI is easy, eliciting maximum capability from current models is harder. She notes the rapid evolution of the Product Manager role, stressing that fast iteration and adaptability are paramount for AI-native products.

The host introduces Cat Wu as the Head of Product for Cloud Code and Co-work at Anthropic, positioning her at the epicenter of AI, product, and building. Cat elaborates on her role, working alongside Boris, the visionary tech lead. She focuses on translating Boris's long-term vision into actionable, cross-functional plans, ensuring all teams are rowing in the same direction.

Cat discusses Anthropic's incredibly fast shipping pace, something the host has never seen before. She attributes this speed not primarily to advanced models like 'Mythos,' but to a deliberate culture of minimal process and removing every barrier to shipping. The goal is to empower every team member to take an idea from concept to launch in less than a week.

Cat explains that traditional product development, with its long planning horizons and emphasis on extensive coordination, is outdated in the AI era. With rapid model improvements, product timelines have shrunk from months to days. PMs now need to prioritize speed, quickly identifying the fastest way to get features out and gathering immediate user feedback, rather than adhering to rigid multi-quarter roadmaps.

Cat outlines two crucial strategies for rapid development: setting clear goals and establishing repeatable processes. Clear goals, like 'professional developers safely get to zero-permission prompts,' eliminate ambiguity and guide decision-making. For process, Anthropic ships most features in 'research preview,' clearly labeling them as early ideas to gather feedback quickly without long-term commitment.

A recent leak of Cloud Code's source code is addressed. Cat clarifies that it was a human error during a package release update, which had passed through two layers of human review. The incident reinforced the need for continuous learning and adding more safeguards, confirming that the individual involved is still with Anthropic and the processes have been hardened to prevent future occurrences.

Anthropic made the difficult decision to deprioritize third-party products using their Open Claude API in favor of first-party offerings. Cat explains this was due to overwhelming demand for Claude and the need to scale infrastructure while making their harness more token-efficient. The move, while upsetting to some, was framed as a necessary step to prioritize their core products and ensure long-term sustainability.

Anthropic employs around 30-40 PMs across several teams. The Research PM team focuses on customer feedback for models. The Core Developer Platform team maintains APIs. Cloud Code focuses on developer tools, while Enterprise ensures adoption for large customers. Finally, the Growth team drives product expansion. This specialized structure allows for focused effort across different product areas.

Cat believes that in the AI era, all roles are merging, with PMs, engineers, and designers often overlapping. Anthropic prioritizes hiring engineers with 'great product taste' to minimize overhead and accelerate shipping. The most crucial skill, she argues, is deciding *what* to build—identifying valuable problems and crafting delightful user experiences. Her own engineering background, like most PMs on her team, fuels this approach.

To navigate the 'tornado' of constant change in AI, Anthropic seeks individuals who 'lean into the chaos.' They embrace challenges with optimism, understanding that stressing over every risk leads to burnout. The team values those who can adapt, wear many hats, and prioritize ruthlessly, focusing on the most important tasks to keep the team moving fast, embodying a low-ego approach to work.

In the quest for speed, Anthropic sometimes sacrifices product consistency. Unlike the past where expensive code led to meticulously planned, single-purpose products, AI's rapid iteration means features might overlap. This can confuse new users about the 'best path' to accomplish a goal. Cat acknowledges the need for better education and onboarding to guide users through their expanding product suite, as rapid launches also mean users struggle to keep up.

Anthropic initially resisted explicit onboarding, believing products should be intuitive. However, the sheer volume of features and user demand for guidance led to tools like '/power-up'. This feature walks users through best practices and cool ways to use Cloud Code, helping them navigate the abundance of features without feeling overwhelmed. It represents a shift from their original 'no onboarding' principle to better serve their growing user base.

Cat identifies two key ingredients for Anthropic's 'otherworldly' growth: a unifying mission and extreme focus. Their mission, to bring safe AGI to humanity, transcends individual product lines, enabling swift, unified decisions. Everyone on the team aligns with this mission, even if it means sacrificing individual product goals. This unwavering focus and mission-driven execution are central to their ability to outcompete larger, more established players.

Cat clarifies the optimal use cases for Anthropic's tools: Cloud Code in the terminal for one-off coding tasks with the latest features, Desktop for graphical front-end work and real-time web app previews, and Web/Mobile for kicking off tasks on the go. Co-work is designed for non-code outputs like creating slide decks, documents, or managing communication, by integrating with various data sources.

Cat describes how Co-work automates tedious tasks for PMs, like creating slide decks for customer meetings. By connecting to various data sources, Co-work can synthesize information, draft presentations, and even tailor content to specific customer needs in minutes instead of hours. This frees PMs to focus on higher-value activities like refining demos, making it a powerful tool for accelerating non-coding work.

Beyond engineering, Anthropic's Applied AI team is the second-largest token spender. This team partners with customers to adopt Anthropic's APIs and model features, often building prototypes and managing extensive customer communications. Their technical expertise combined with customer-facing responsibilities makes them heavy users of both Co-work and Cloud Code, driving innovation in custom work software for specific use cases.