Product SideQuest: a practical typology of AI PM roles and how to prepare for each

Diego Granados has led AI product experiences at Google — including the Data Science Agent for enterprise users — and built ML models at Microsoft. His March 2026 guide on becoming an AI product manager is notable for what it doesn’t do: it avoids the generic “learn the basics of ML” framing and instead distinguishes three roles that all carry the AI PM label but require substantially different skills and preparation.

The first type is the AI Experiences PM. This role designs user-facing products powered by AI — a chat assistant, a recommendation system, a document summarizer — and its core challenge is trust and usability rather than model construction. An Experiences PM decides how the product communicates uncertainty, what the interface does when the model fails, and whether the design calibrates user expectations appropriately about what AI can and cannot do. Granados introduces the concept of “AI product sense” — the practiced judgment needed to recognize when probabilistic output is good enough to ship versus when its failure modes make it unsuitable for a particular user context. This is the role most often discussed in AI PM articles, and also the most misunderstood, because it looks like traditional product management until you run into its AI-specific failure modes.

The second type is the AI Builder PM. This role manages platforms, infrastructure, and models used by internal engineering teams: the fine-tuning pipeline, the evaluation framework, the embedding infrastructure. Builder PMs work far closer to engineering and require deeper technical fluency — understanding of model evaluation metrics, experiment design for ML systems, and the trade-offs between latency, cost, and accuracy. Granados draws a sharp line here: the Experiences role and the Builder role share almost no day-to-day skill overlap, despite appearing in the same job title category.

The third type is the AI Enhanced PM — not a destination but a baseline. This is the use of AI tools to do PM work faster: writing PRDs with LLM assistance, summarizing customer interviews, generating competitive analysis. Granados argues this competency is now required across all PM roles and should not be confused with specialization in building AI products. It is table stakes, not a differentiator.

The guide draws on actual postings from Airbnb, Meta, Bumble, Spotify, and Stripe, with compensation figures ($155K–$280K+ for senior roles) and specific technical requirements. This grounding in market data is one of the article’s more useful features. Rather than describing an idealized future role, it describes what companies are actively hiring for.

The practical advice for career transition centers on building and shipping something publicly — using tools like Lovable or Claude Code — as the most reliable way to develop both credibility and judgment simultaneously. Granados is skeptical of certification programs as primary preparation paths; his view is that direct product experience with AI systems is what distinguishes candidates in interviews.

Most useful for product managers evaluating which direction to move when transitioning into AI-focused roles, and for those trying to understand what distinguishes the different types of AI PM work in practice.