HelloPM: AI product management masterclass 2026

What the video covers

Published in January 2026, this is a full-length masterclass by Ankit Shukla, founder of HelloPM and a product and growth professional with over ten years of experience. The session ran live with over 800 attendees and is available on YouTube as a replay. It covers the technical foundations and applied knowledge that product managers are expected to bring to AI product roles.

The session is structured around what an AI-focused PM actually needs to understand: not just how to use AI tools, but how the underlying systems work and why. Shukla walks through how large language models generate output through next-token prediction, what the transformer architecture does, and how tokenization affects prompt design in practice. This is followed by a GenAI value stack — a four-layer framework for understanding where value is created in AI products, from infrastructure through model providers through application layer through interface.

The technical content then moves to retrieval-augmented generation (RAG): how RAG pipelines work, what embeddings are and how vector databases store them, and why RAG is typically preferable to fine-tuning for most enterprise product use cases. Prompt engineering frameworks follow, with an emphasis on context engineering — structuring the information provided to a model to improve output quality consistently rather than relying on phrasing.

The session closes with AI agents and guardrails, then three product teardowns: Granola (AI meeting notes), NotebookLM (AI document research), and Lovable (AI-assisted product building). The teardowns apply the technical concepts to real products, showing how design decisions connect to model behavior and where each product’s tradeoffs are visible.

Who it’s for

Product managers who have been using AI tools casually but lack a clear mental model of how they work. The session is pitched at beginner to intermediate level — no programming experience is required, but participants with some prior familiarity with LLMs will get more from the teardowns. PMs in roles that involve specifying AI features, writing AI acceptance criteria, or working closely with ML engineers will find the foundational portion particularly useful.

Key takeaways

1. Context engineering matters more than prompt phrasing. Shukla’s framing of the LLM interaction model emphasizes that well-structured context — what you give the model to work with — has a larger effect on output quality than the wording of the instruction itself. Understanding this changes how a PM writes feature specifications for AI-powered products.

2. RAG versus fine-tuning is a product decision, not just a technical one. The choice between retrieval-based approaches and fine-tuning has cost, maintenance, and update-frequency implications that PMs should be able to reason about with engineers, rather than deferring entirely to technical teams.

3. The four-layer GenAI value stack clarifies where competition happens. Separating infrastructure, models, applications, and interfaces helps PMs identify which layer their product actually competes on — and where defensibility comes from.

4. Agent guardrails are a product design problem. The section on AI agents treats boundary conditions — what the agent should refuse, escalate, or confirm with the user — as product requirements, not just safety concerns. Defining these boundaries is framed as one of the central responsibilities of an AI PM.

5. Product teardowns reveal design tradeoffs embedded in real products. The Granola, NotebookLM, and Lovable teardowns show how understanding the underlying architecture reveals why each product makes the design decisions it does, and where the next iteration could improve.

Worth watching if…

You are a PM who wants to have more precise technical conversations with ML engineers, or if your team is in early planning for an AI product and needs a shared vocabulary for architecture decisions. Also useful as preparation before taking a paid AI PM course, since the session covers the foundational vocabulary that more advanced programs typically assume you already have.