Curriculum

Prototyping AI Products With Coding Tools

PM-as-builder with agentic coding tools is a named gap and a named frontier. The report lists Claude Code (16), Cursor (14), Codex (5), CLI (11), GitHub (11), APIs (24), TypeScript (21), and Python (14) appearing in PM postings, and says th

1. 03.01

   Build and iterate in Claude Code

   You stand up and change a working agent yourself inside an agentic coding tool, so the spec you later write is one you have already put your hands on.
2. 03.02

   Read the code you did not write

   You open an agent someone or a model already built, point to its loop, its tool calls, and the settings that decide its behavior, and read a pull request as a behavior change.
3. 03.03

   MCP first contact

   You wire an agent to a single tool over MCP and read the resulting trace, so the protocol that reaches enterprise systems is something you have run, not just named.
4. 03.04

   Prototype by what it must prove

   You pick the prototype by the single uncertainty it has to kill, and load the real constraints into the first prompt, so one pass gives you a usable signal instead of a charming demo.
5. 03.05

   One output is an example; ten inputs by five r

   You judge AI features by running varied inputs five times each against a bar set in advance, then use the scores to choose go, iterate, or kill and scope v1 to what is reliable.
6. 03.06

   INTEGRATION CASE

   From one named risk, you can turn evidence into a test, a spec with a pass rate, and a plan with cuts.