[joshvince]

You're among the first people I've encountered with both experience in writing your own software and prompting an LLM to do it. I have a few questions for you (maybe the answers will make you feel better, or maybe they'll make you feel worse)

- when the code the model spat out was wrong, how did you fix it? Did you identify it was wrong before you ran it?

- what level of complexity was there in the code, in terms of "business logic" or complexity of the requirements you fed into it?

I ask these two questions because I am not sure an AI will get to the level of experience you have in the near future in _generating_ complex applications, let alone being able to reflect on why its own creations are wrong, fixing them, deploying some output, and then explaining the changes?

A human is going to be in the loop in these cases for a long time, and I'm assuming part or all of your decades of experience has been spent understanding quite how poor people are at explaining their requirements. What if you thought of generative AI as a tool you can learn to utilise to do your job more effectively?