A big, vague request like "build the whole comment system" tends to produce a big, hard-to-review result. If something is wrong, it is buried somewhere in a giant diff. The fix is to break the work into small steps you can verify one at a time.
Start by getting the shape right before any code is written. Plan mode is built for this. Cycle to it with Shift+Tab, and Claude explores and proposes a plan without changing anything. You read the plan, correct the parts it got wrong, and only then let it implement.
Once you have a plan, tackle it in checkable chunks:
1. Add the Comment model and its schema. Show me before moving on.
2. Add the POST /api/comments route with validation. Run the lint check.
3. Add the admin moderation endpoint. Test it against a sample comment.
4. Wire up the comment form component.
After each step, you confirm it works before the next one starts. When step two breaks, you know exactly where the problem is, because everything before it was already verified.
Why this beats one giant request:
- Errors surface early. A mistake in step one is caught before it is built on.
- Reviews stay small. A focused diff is easy to read; a sprawling one hides bugs.
- You keep control. You can adjust direction at each checkpoint instead of hoping the whole thing lands.
Pair each step with a concrete check, a test, a lint run, a manual look, so "done" means verified, not just written. This is the Explore, Plan, Implement, Verify loop in practice.
The instinct to ask for everything at once is understandable, but smaller steps almost always finish faster overall, because you spend less time untangling a large result that went wrong somewhere in the middle.
Comments
Be the first to comment.