We improve step by step, not all at once

The finished result is almost never born from the first request - it is grown step by step, in small repetitions. There is one rule: one step - one change. When you change five things at once and something breaks, it’s not clear which of the five is to blame; changing one at a time, you always know the reason. It's like adding salt to soup in pinches, tasting, rather than pouring out a packet at once. Describe the current state (what is already there and working), one specific change and, of course, what to leave untouched. The phrase “don’t change the rest” saves hours: without it, the AI ​​often rewrites the entire code again, and working pieces suddenly break. After each edit, ask for a short summary of “what has changed” - this way you keep track of history and can roll back if the new version is worse. A common mistake is to accumulate ten wishes and dump them all at once, and then not understand why everything fell apart. Small steps are slower in appearance, but faster in reality - fewer rollbacks.