For decades, product managers operated within a relatively stable world. Their job was to understand customer problems, identify worthwhile opportunities, align engineering efforts, and make thoughtful tradeoffs between competing priorities. Success depended on building products that solved real problems while balancing technical complexity, business goals, and user experience. Generative AI introduces products whose behavior cannot be described entirely through predefined workflows or deterministic rules. Models respond differently depending on context, reveal new capabilities as users experiment with them, and occasionally fail in ways that neither engineers nor researchers anticipated during development. A feature can appear remarkably useful in one situation while creating confusion or misplaced confidence in another. That means product managers increasingly spend their time thinking about questions that barely existed a decade ago. Beyond deciding what deserves to be built, they also have to consider how people will interpret the system’s behavior, what level of confidence users should place in its responses, how uncertainty ought to be communicated, and when a technically impressive capability is mature enough to become part of an everyday product. Reading about Anthropic’s philosophy reinforced that impression. The company consistently describes its mission in terms of building AI systems that are useful, honest, and harmless while investing heavily in research, evaluations, alignment, and responsible deployment. Those priorities suggest that product decisions cannot be separated from questions of reliability and trust. Shipping another feature may increase the model’s usefulness, but it can also change how people depend on the product, how they interpret its answers, and what they expect it to do next. That creates a product management discipline that feels noticeably different from the one many software teams have practiced for years. Perhaps the defining...