Asking for One Change at a Time

You have a working quiz app. You want to add a timer, change the color scheme, fix a bug where the score is not updating, and add a progress bar. So you write one prompt asking for all four at once. The AI responds with a massive block of code. You paste it in. Three things work, one does not, and now you are not sure which part of the new code caused the problem. This is one of the most common frustrations in vibe coding — and it is almost entirely self-inflicted. The fix is simple and powerful: ask for one change at a time.

Why Focused Requests Win

When you ask for a single, focused change, several good things happen: You know exactly what changed. If something breaks after a focused request, the problem is almost certainly in that one change. Diagnosis becomes straightforward. The AI stays in scope. When you give the AI multiple tasks at once, it has to juggle them — and it sometimes drops one, merges two together, or introduces a solution to task B that conflicts with task A. One task at a time means full attention on one outcome. You can verify before adding more. Confirm the timer works before you ask for the color change. Confirm the color change looks right before you add the progress bar. Each successful verification is a checkpoint you can return to if something later goes wrong. You build confidence in the codebase. Incremental changes you understand and have tested feel solid. Sweeping rewrites you accepted without verifying feel fragile.

Here is a concrete before-and-after. You have a working to-do list and you want three improvements. Unfocused prompt: 'Make the to-do items draggable so you can reorder them, add a filter button that shows only incomplete items, and change the delete button from an X to a trash-can icon.' Focused approach — three separate prompts: Prompt 1: 'Add a filter button above the list. When clicked, it should toggle between showing all items and showing only incomplete items. The button label should update accordingly — All Items or Incomplete Items.' [Verify it works.] Prompt 2: 'Change the delete button icon from an X character to a trash-can emoji. Nothing else about the delete behavior should change.' [Verify it works.] Prompt 3: 'Make the to-do items draggable so the user can reorder them by dragging and dropping.' [Verify it works.] Three prompts. Three clear checkpoints. If the drag-and-drop breaks something, you know it happened in prompt 3, not somewhere in the tangle of one big response.

When to Bundle and When to Split

Splitting every change into its own prompt can feel slow. There are cases where bundling makes sense: Bundle changes that are tightly coupled. If you need a new data field and you need to display that field in the UI, those two things are so connected that splitting them would produce broken intermediate states. Bundle trivial style changes. Changing three button colors at once is genuinely one logical task — 'update the color scheme' — and splitting it into three prompts just adds noise. Split changes that are independent. A timer and a color scheme have nothing to do with each other. A bug fix and a new feature have nothing to do with each other. These belong in separate prompts. The question to ask yourself: 'If the AI only does one of these things, does the result still make sense?' If yes, split them. If one without the other would leave things broken, bundle them.