Introduction: AI will widen the gap between PMs, not steal their jobs
Generative AI has become commonplace, and the work of PMs (Product Managers) has changed dramatically. Specifications, requirements refinement, research, UI copy drafts, and backlog shaping—such tasks can be done faster by AI. But to that extent, the quality of decisions about what to build, why to build, and how to win will be more important than ever for product success.
This article summarizes the 10 essential skills for thriving PMs in the AI era in a friendly, approachable way that you can try starting tomorrow.
1. Problem Framing: Decide what to solve before feeding it to AI
The more you use AI, the more important the way you frame the problem becomes. Ambiguous questions yield ambiguous answers. A PM's value shows up in the initial way you frame the problem.
- Bad example: I want to reduce churn. Give me some initiative ideas.
- Good example: The Time to Value (TTV) for new users is long. Where in onboarding do users drop off? I want to split into three hypotheses and create a validation plan.
The trick is to pass AI a set of background, constraints, success metrics, and target users.
2. AI Literacy (Assessing capabilities and limitations): Manage expectations
PMs should at minimum understand that LLMs generate plausible text but do not guarantee truth. Hallucinations and biases from training data are realities in practice.
Additionally, multimodal AI handling text, images, audio, and video has become standard, making configurations like Retrieval-Augmented Generation (RAG) and agents increasingly common. Even if you can't implement everything, being able to explain the concepts conceptually greatly raises the resolution of conversations.
3. Data Thinking (Metrics design and measurement): AI products won’t grow unless you can measure them
Because AI outputs can vary, defining what constitutes “good” is harder than before. Therefore, PMs must own metrics (KPIs) and evaluation design.
- Product KPIs: Retention, churn, NPS, TTV, feature usage, etc.
- AI quality metrics: Accuracy, usefulness, reproducibility, toxicity, cost ($ / request), latency (response time)
- Operational metrics: Failure rate, escalation rate, human review ratio
Practically, link analytics platforms (Amplitude / Mixpanel, BigQuery, Looker, PostHog) with evaluation logs (prompt, context, output, user reaction) to sustain a continuous improvement loop.
