AI coding partners.

Lessons in this module CORE + RECIPE

Work through these in order. Each lesson ends with a project checkpoint that adds to your directed-edit log draft; Lesson 3.5 freezes the log.

1. 3.1   What a coding agent actually is.

   Mostly CORE

   The mental model that separates a suggester from an agent. The four moves of a directed edit — locate, plan, write, verify — with the director’s check laid over the top. Why cloud models are the default and when to use the local model instead. Why the Code tab in the Claude desktop app is Module 3's home base, and how it differs from the Cowork tab (which shows up in Module 4 and later). Pairs with the Anatomy of a directed edit worksheet.

   Open Lesson 3.1 →

2. 3.2   Your first directed edit.

   Mostly RECIPE

   Download the tidy-starter practice project and direct two real bug fixes end to end in the Code tab. Optional advanced sidebar: do the second bug in the Claude Code CLI for a side-by-side comparison if you set up the CLI in Lesson 2.4. Both walkthroughs ship with safe-default and troubleshooting sub-callouts. Ends by writing Entry 1 of your directed-edit log and copying the two safety norms to the top.

   Open Lesson 3.2 →

3. 3.3   Reviewing what the agent did.

   Mostly CORE

   The most important lesson in Module 3. The four ways an agent's edit can go wrong (misplaced change, scope creep, silent deletion, plausible wrong), the five-move review you run on every change (check the shape, read the agent's summary, ask the agent, run the result, iterate), and the written checklist you use for your first thirty directed edits. Built for directors, not programmers — you don't need to read code to do this. The Diff Review Trainer gives you three real scenarios to practice on, two of which are built to fool you.

   Open Lesson 3.3 →

4. 3.4   When AI coding works brilliantly and when it fails.

   Mostly CORE

   The most CORE-heavy lesson in Module 3. Four strong zones where a directed coding agent genuinely outperforms solo programming. Eight tripwires where the same tool confidently fails, split into properties of the task vs. properties of the session. Three reframe habits (make goals legible, prefer additions to rewrites, build the verify signal) that bend tripwire work back toward the strong zone. Plus the zones map you keep on your wall.

   Open Lesson 3.4 →

5. 3.5   Scoping larger changes.

   CORE + RECIPE

   Features, not bug fixes. Write a scoping doc — goal, what the change touches, smaller pieces called slices, how you'll know it worked, and what's explicitly off-limits — and direct an agent through the slices using the scaffold → path-through → polish pattern, resetting context between slices and saving your work after each. Write Entry 3 of the log and freeze /capstone/directed-edit-log-v1.md — the third capstone artifact of the course.

   Open Lesson 3.5 →

---

Resources for this module RECIPE

Six printable companions and one interactive activity run alongside the lessons. Each printable is a print-to-PDF page that prints cleanly on letter paper. Use them in the order their lessons reference them.

• Anatomy of a directed edit — six-task printable worksheet for Lesson 3.1. Labels the four moves, adds the director’s-check lane, and ends with a reflection line for the student’s own most-recent coding-agent use.
• Directed-edit log template — the template that grows across Lessons 3.2, 3.3, and 3.5 into your directed-edit-log-draft.md. Three entry blocks, the two safety norms at the top, and a zone-map section you fill in at 3.4.
• Diff Review Trainer — interactive HTML activity for Lesson 3.3. Three staged scenarios — one obviously fine, one with scope creep, one plausibly wrong with passing tests — each presented as the agent's plain-English summary plus the diff shape and test output (no code-reading required), with clickable follow-up questions and reveal-on-click scoring.
• Director's review checklist — reference card for Lesson 3.3 and every subsequent module. The five-move review on one side, the four failure modes on the reverse. Designed to tape to the side of your monitor.
• Tripwire catalog — eight-row worksheet for Lesson 3.4. Each tripwire gets a one-sentence definition, a specific example, and a counter-move. Ends with eight worked zone-predictions and space for your own ongoing additions.
• Scoping doc template — fillable for Lesson 3.5. Five sections — goal, surface area, slices, verify signals, out-of-scope list — with a sidebar reminder of the scaffold → path-through → polish pattern.

---

What you should have when this module is done CORE

By the time you close out Module 3, you should be able to point to six concrete things on your machine and in your capstone folder:

• A cloned starter repo (or your own codebase) with at least two merged bug-fix commits and one merged feature-sized change, each committed in slices.
• A saved scoping doc in the repo you worked on, showing goal / surface area / slices / verify signals / out-of-scope list.
• Your personal tripwire catalog (tripwire-catalog-v1.md) with an example and counter-move for each of the eight tripwires.
• Your diff-review checklist printed and in sight of your editor. If it is not physically visible to you, the habit has not landed yet.
• A frozen /capstone/directed-edit-log-v1.md containing, in order: the two safety norms, the zone map, and three entries (Code-tab bug fix, deliberately-hard review, feature-sized slice sequence).
• The completed end-of-module check in your portfolio, scored at 11.5 / 15 or better with full credit on at least one applied question.

If any of these is missing, go back to the checkpoint that produced it and finish before moving on. Module 4 assumes the directing muscle is real — the research agents it teaches are the same loop on work that is not code.