PMCurve: Product management and AI in 2026

Published in December 2025, this video is from Deepak Singh, founder of PMCurve — a platform focused on technical product management education. Singh runs the Advanced AI Product Management program on Maven and has spent several years helping senior engineers and PMs build the skills needed to lead AI product work. The video is a standalone piece, not a course module, and is aimed at PMs who already have baseline familiarity with AI tools and are now trying to think more systematically about their product work.

Who it is for

The video addresses PMs who are past the introductory stage — already using AI tools or building AI features — and are now navigating the harder questions about evaluation, governance, and adoption at scale. It covers strategic and evaluative thinking rather than tooling basics, making it more relevant for mid-level and senior PMs than for those just starting out with AI.

Key takeaways

1. Evaluating AI opportunities requires a different lens. Standard prioritisation frameworks do not transfer cleanly to AI features because the effort side is poorly defined when model performance is variable and continues to improve. Singh makes the case for explicit evaluation criteria specific to AI opportunities.

2. Governance is becoming a product concern, not just a legal one. As AI features move from internal tools to customer-facing experiences, questions about how AI decisions are made and how they can be reviewed or overridden become part of the product surface area. Singh treats this as a design and specification problem for PMs.

3. Building for trust is distinct from building for usability. AI features frequently pass usability testing but stall at adoption because users do not trust the outputs enough to rely on them in consequential situations. Singh argues that designing for trust means making AI outputs interpretable and giving users meaningful ways to verify or correct them.

4. Discovery for AI products needs to include model failure modes. The discovery phase for an AI feature should include explicit research into where the model currently fails, not only what users want. Building around known failure modes before launch is more effective than discovering them in production.

5. The 2026 coordination problem. Singh’s assessment is that individual AI tools are now capable enough for many PM tasks, and the harder problem is coordinating AI work across a product team — consistent prompting practices, shared context, and clear ownership of AI outputs. This is a team and process design problem as much as a tooling one.

Worth watching if

This video is most useful for PMs in product organisations who are starting to think about AI not as an individual productivity tool but as a team capability that needs structure and governance. The 10-question framework Singh presents is a solid starting point for a team planning session, a roadmap review, or an internal audit of how AI is currently being used across a product function.