Key Takeaways

  • Anthropic's Head of Product for Claude Code, Cat Wu, says "product taste" is now the most valuable skill for PMs building AI, as the cost of writing code drops to near zero.
  • Defining product direction even a month out has become extremely difficult in AI. PMs must deeply understand current model capabilities and predict how user behavior will shift with new AI.
  • Daily feature shipping at Anthropic is common, driven by a clear company mission and direct model capabilities, shifting the PM role from specification to rapid iteration and gap-filling.
  • Human common sense and emotional intelligence remain irreplaceable. AI models lack these, meaning PMs still need to provide intuitive understanding and adapt to amorphous roles.

"What to Build" Beats "How to Build"

The ground has shifted under product managers in AI. Cat Wu, who leads product for Claude Code at Anthropic, makes this point sharply. She says the focus isn't on shipping code anymore, because writing code is becoming cheap. Wu states, “As code becomes much cheaper to write, the thing that becomes more valuable is deciding what to write.”

This means the product manager's core value isn't knowing how to implement but knowing what to implement. This is where "product taste" enters the picture. Wu argues it's the single most important skill. It’s the ability to discern what users truly need, what feels right, and what will resonate, even before robust data exists. "I still think it comes back to product taste... I think that that skill set can come from any background but I think that's the most important thing," Wu explains. In a world where AI can generate countless options, the human element of selection, refinement, and intuition becomes paramount. Building bad features quickly with AI is just as easy as building good ones; product taste helps differentiate.

Navigating the Month-Ahead Fog

Traditional product roadmapping often looked 3, 6, or even 12 months out. That’s a fantasy in AI, according to Wu. She admits, “I think the hardest skill is being able to define what the product should look like a month from now.” The capabilities of foundational models change so rapidly that a feature concept viable today might be obsolete or easily replicated tomorrow.

This constant flux demands a new kind of PM. One who doesn't just understand user problems, but also deeply grasps "the current model." Wu advises PMs to understand "how do you elicit the maximum capability?" from the current AI tools. This isn't about knowing every API endpoint, but about recognizing the bleeding edge of what's possible and how user interaction patterns will evolve. It requires continuous learning, experimentation, and a willingness to scrap plans that felt solid just weeks ago. For Anthropic, their mission guides this rapid iteration, allowing them to ship features daily, adapting on the fly rather than sticking to rigid, long-term specs.

The Enduring Human Edge: Common Sense and EQ

Despite the rapid advancements of AI, Wu stresses that humans still offer something AI models don't: common sense and emotional intelligence. She notes, “I think humans still provide a level of common sense that the models don't.” This isn't about AI limitations in a philosophical sense, but practical gaps in product development. AI can optimize, generate, and process, but it can't intuitively understand nuanced human needs, social cues, or the unstated context of a user’s interaction.

As AI takes over more rote tasks, the PM role becomes more amorphous. Wu talks about being "adaptable to fill gaps." This means stepping into whatever role is needed—whether it’s user research, competitive analysis, or even directly shaping prompts—to ensure the AI product truly serves its users. This requires a high degree of emotional intelligence, not just for team collaboration, but for truly empathizing with users and translating their unspoken needs into product direction.

What to Do With This

Stop chasing the latest AI coding framework. Instead, spend 5 hours this week deeply engaging with your product's target users—call them, shadow them, or simply read their forum posts—to refine your own "product taste." Then, take your current 3-month product roadmap and cut it to 3 weeks, forcing yourself and your team to focus on the immediate, model-driven opportunities.