Last week you generated an app from a brief. That works beautifully right up to the moment you need a change the prompt-to-app tool cannot cleanly make, a specific fix, a bit of custom logic, a feature that touches several parts of the app at once. This week you cross that line, and the mindset you cross it with matters more than any tool.

The shift is this: you are not learning to code, you are learning to direct code and verify it. A coding agent like Claude Code can read your whole project, make a change across several files, and explain what it did. Your job is not to write those lines. It is to say clearly what you want, and then to check that what came back is right and did not break anything else. That second half, verifying, is the part beginners skip and the part that actually protects you.

So learn the safe loop now, because it is the whole game and it never changes. Plan first: before any code changes, have the agent lay out what it intends to do, and read that plan. Make one change: keep each instruction small and scoped, one feature or one fix, not five at once. Review the diff: a diff is just a before-and-after showing exactly what changed; you read it to confirm the change is what you asked for and nothing else moved. Keep or roll back: if it is right, keep it; if it is wrong, undo it and re-describe. Plan, change, review, keep-or-roll-back. Do that and you can safely direct changes to software you could not have written, which is the entire promise of this week.

Why small and scoped? Because when a change is small, a wrong result is obvious and easy to undo. When you ask for five things at once and something breaks, you cannot tell which of the five did it. Scoping is not a limitation, it is how you stay in control of a system you are directing rather than writing.

By the end of this lesson you have the working method. The next lesson points it at your actual app.