I use ChatGPT and Claude Code a lot in my programming now. I used to start by stating the problem or goal and then letting it write the code immediately. But I ended up iterating a lot because AI would make many incorrect assumptions.

The number one thing that has improved the quality of the code is when I started saying this at the end of the prompt: Please ask me questions one at a time before coding.

By having ChatGPT ask questions, it is far less likely to make assumptions. At the same time, it surfaces questions that I should have thought about when describing the task. So the result is much cleaner code. The “one at a time” part is simply to get ChatGPT to not ask a bunch of questions all at once. Thinking about each question individually makes for a better process.

The “one question at a time” approach is slower up front. It forces you to take time to work through how the code is best constructed. In the end, I find that I’m saving a lot of debugging time. Nothing about this is really new, it’s just coding best practice. What’s new is how to prompt, and it’s worth making a ChatGPT session into a real conversation. I find that results in better code.