I agree. I can get behind people using AI to build/replace stuff, but if it's anything beyond stupidly simple it's going to require a lot of thought and time - even if that ends up being time spent writing increasingly detailed prompts and testing the output instead of writing code directly.
>The idea that you can write one sentence and get a masterpiece of software that meets your needs is delusional and I’m shocked that it gets as much currency as it does here.
Remember Z language? Or UML? Languages that were supposed to be able to generate a perfect piece of software based on a perfect spec?
Yeah ... those went nowhere :)
I asked Copilot to tell me how to clear out a a RabbitMQ queue and it gave me a great answer. I asked it to write me a Python script that solves Tower of Hanoi, same thing.
In both cases it is a well-defined task and I don’t care about how it is done.
Applications software is different. If you farm out a task to offshore developers you are either going to have to document every. little. detail. (India) or get a spike of work which might get between 0-40% of the way there and come back with detailed feedback through several cycles (China)
Maybe an A.I. coder could work with you intensively to get your needs met, but it’s not going to be “implement user authentication” and then it is done unless your definition of done is launching and getting a call the next night that somebody logged in without a password.
Even so, AI does make people more productive and it will shift buy vs build discussions towards build.