The loop in depth

Module 3 names the durable AI-coding loop end-to-end: intent → ask → evaluate → steer. Four lessons unpack the loop, one per step. Every M3 lesson runs the same worked example on a single self-contained scratch/index.html (per D-31), shown in parallel on both Claude Code AND Gemini CLI (per D-25 + D-27) so the loop is taught as durable across agents. You run only the agent you picked in Module 0 (per D-29); the other agent's transcript is shown as reference.

Loop checks across the four lessons name intent, ask, evaluate, steer — one per lesson, in order (per D-24).

What this module builds

By the end of this module, you can run the durable AI-coding loop — intent → ask → evaluate → steer — on any agent: knowing what you want, asking for it well, judging whether the output matches, and course-correcting when it doesn't.

Each lesson builds on the last:

• Lesson 1 — Introducing the loop: name the four loop steps in order and run one complete iteration on the scratch starter → sets up Lesson 2 by leaving the ask step as the one to sharpen first.
• Lesson 2 — Planning vs execution: tell a "plan it, don't build it yet" ask apart from a "now do it" ask, and use four slash commands to keep a session's working memory in check → sets up Lesson 3 by giving you a plan to compare the agent's actual output against.
• Lesson 3 — Reading plans, recognizing wrong: run five observation patterns to spot wrong output — including a hallucinated answer — without reading the code → sets up Lesson 4 by leaving you holding a wrong result you now need to fix.
• Lesson 4 — Steering and recovery: write a three-part steer that course-corrects the agent, catch it over-engineering an open-ended ask, and know when /clear and a fresh start beats another steer → sets up Module 3.5, where you add the code-reading floor on top of this observation-only loop.

The thread that ties it together: Module 3 is about the durable loop — intent → ask → evaluate → steer. The keystrokes change as agents evolve; the four steps do not. Name them once, on two different agents, and the skill carries to whatever agent comes next.

Worked example

All four M3 lessons progress a single self-contained scratch/index.html across iterations (per D-31). The starter ships in Plan 02-02; M3 L4 ends with: "you can delete the scratch directory now — your real project starts in Module 4."