Plan, build, review

The biggest mistakes with coding agents happen before any code is written: the agent heads off in the wrong direction and you find out 400 lines later. The fix is a rhythm Anthropic teaches in Claude Code 101: explore, plan, code, commit.

1. Explore. Let it read the relevant files first, without changing anything.

2. Plan. It proposes the change in steps. You read the plan and correct it. This is the cheapest moment to catch a bad idea.

3. Code. It makes the change, small enough that you can review it.

4. Commit. Once the checks pass, save it with a clear message.

Plan mode does steps 1 and 2 for you: press Shift+Tab until you see it, or start a single prompt with /plan. Claude reads and proposes, and can't edit anything until you approve.

Keep each change small. One feature or one fix per task. A small diff is one you'll actually read; a giant one you'll wave through. Between unrelated tasks, run /clear so old context doesn't leak into the new job.

Tests are the agent's eyes. An agent can't look at your app the way you do. A test, a typecheck or a linter gives it a yes or no it can see. Ask it to run the check and show you the output, not just say "done". Matt Pocock goes further: have the agent write the failing test first, so the fix has a target.

The left reply sounds productive, but you now have a sprawling diff and no proof. The right one did less, told you what it touched, and showed the test run.

Your turn. Brief an agent to add that "closed today" banner. Put the homepage in src/pages/home.tsx and the hours logic in src/hours.ts. Your brief should keep the scope tight (say what not to touch), name the check (pnpm test), ask for the test output as proof, and ask for the plan first.