AI product management is more than prompts

Is an AI product manager really just a product manager with a few new tools? Two experiences helped sharpen my answer.

Years ago, I hired a product manager. When we worked together again, AI was central to the product and she described herself as an AI PM. In another team, a new AI PM was dealing with serious hallucination issues within weeks of joining.

Watching their work made the difference tangible.

The “not always” is part of the job

Traditional product management largely deals with deterministic systems: if you build a feature correctly, it should behave as designed.

AI introduces probabilistic behavior. If you design the system well, the model will often produce the result you want. But not always.

That “not always” changes the work. Alongside roadmap, UX, prioritization, and business impact, the product manager needs to manage:

• Trust and accuracy.
• Compliance and predictability.
• Evaluation and feedback loops.
• Latency, cost, and scalability.

Managing behavior as well as features

A product manager asks whether a feature should be built. An AI product manager also has to ask whether its behavior can be trusted enough to ship, scale, and deliver value in real use.

Knowing how to write prompts is useful. Taking responsibility for behavior, outcomes, and risk requires a much broader set of judgments.

That’s the part of AI product management I find most important—and most demanding.