> Otherwise it’s all a black box. You ask for something, code comes out, and you hope it does what you meant.

It's easy to argue against the worst example of something. Using AI to program can be done well, or it can be done poorly. Using compilers and interpreters to program can be done well or poorly as well. There is nothing fundamental about AI programming that limits programmer comprehension, care, or influence over what they create.

There are different disciplinary techniques that can be applied in order to achieve this (rather than blindly promoting and hoping for the best). I have personally been leaning into spec driven development, which I have found success with. I use a tool called openspec to drive the framework of creating proposal, design, and task documents, ahead of driving the implementation. I will use it to spend sometimes up to hours iterating on specifications, until I reach something that fully outlines what I want. I read every word of the specification files to make sure they fully align with my understanding of the problem space. Once I'm happy, I can confidently tell the agent to implement the specification, walk away, and come back to the specification being implemented.

I care about the software I create. Using AI to develop does not get in the way of that.