I never start with a blank slate. I know what I want to accomplish, and I usually have some idea of how I want it done. Sometimes that is a rough direction. Sometimes I already have a specific approach in mind.
That doesn’t mean I write a long brief every time. Most of the time, I can make a straightforward request without supplying background or examples, and AI picks up what I mean. I’d say that works about 90% of the time for what I ask it to do. From there, we work out the details.
Agree on a plan
I treat AI as a teammate. I bring ideas, it makes suggestions, and we work through them together. Having an approach in mind doesn’t mean I have to stick with it if a better option comes up. Before we start making changes, we agree on a plan. I want to know what we are doing and what to expect next. If a step sounds ambiguous, I ask what it actually involves before we move forward.
Work in small pieces
We always work in small pieces. To be honest, that is usually how AI recommends proceeding anyway. It gives me a chance to follow each change, try it, and ask questions before we add more on top of it.
I read the code as we go. I want to understand how it is put together and whether it introduces any security concerns. If I don’t understand a function or why something was done a particular way, I ask for an explanation. If that explanation is still unclear, I keep asking. I need to understand the process and the code before it is committed.
Get a second opinion
I also use multiple models alongside each other to get a second opinion. I can take an approach or a piece of code to another model and ask it to review it. That gives me another perspective to consider as part of my own review.
Try it before releasing it
I build and test locally first. Reading the code is one part of the review; seeing it run is another. Once the changes have been reviewed, I push them to a preview site and check them again before release.
That is how I am comfortable working with AI. We can move quickly, but I still want to be able to explain what we built and why it works the way it does.