Curriculum

Discovering and Sizing AI Opportunities

AI PM craft is the second strongest theme (88 percent, 73 companies) and building under ambiguity is 68 percent, both core. The report asks for Shreyas-style product sense so the AI PM makes correct macro and micro calls before the data exi

1. 04.01

   Pre-data bets

   Before usage data exists, you simulate one user's path, pick the riskiest assumption, and record each product call as a dated bet you will later retire or revise.
2. 04.02

   Discovery for AI

   You run discovery on an AI product by mining real interviews and production logs for the one behavior you can actually move, not a wishlist of features.
3. 04.03

   The problem-first test

   Before you scope an agent, you apply the problem-first test to decide whether an agent, a plain workflow, or nothing at all is the right answer.
4. 04.04

   Qualify the opportunity

   You test whether an opportunity is hard enough and material enough to deserve an LLM, and when it is not, you name the classifier you should have shipped instead: labeled data, a threshold, and a retraining plan.
5. 04.05

   The capability horizon

   You judge the capability horizon as your strategy spine, run the model-upgrade drill on every frontier release, and use the results to decide what compounds, what to refuse, where rented capability still fails, and whether a real signal loop can keep improving quality.
6. 04.06

   Build, buy, or open weights

   You score building on a model API, buying an agent platform and running open weights in a fixed order, then write the one-page call leadership can approve.
7. 04.07

   Goals and tradeoffs

   You turn strategy into cascading goals, choose work when everything feels urgent with a clear scoring rule, and say no to the loudest request with a priced alternative instead of a flat refusal.
8. 04.08

   INTEGRATION CASE

   Based on source-backed evidence, you draft a falsifiable problem statement and a strategy memo a stranger can defend.