Yes, the best way is to type the real program completely into the AI, so that ClosedAI gets new material to train on, the AI can make some dumb comments but the code works.
And the manager is happy that filthy programmers are "using" AI.
>>However, the AI is hard to work with, it expects specific wording in order to program our code as expected.
Speaking English to make something is one thing, but speaking English to modify something complicated is absolutely something else. And Im pretty sure involves more or less the same effort as writing code itself. Of course regression for this something like this is not for the faint hearted.
Actually I think that's the near future, or close to it.
1. Humans also need specific wording in order to program code that stakeholders expected. A lot of people are laughing at AI because they think getting requirements is a human privilege.
2. On the contrary, I don't think people need to hire AI interfacers. Instead, business stakeholders are way more interested to interface with AI simply because they just want to get things done instead of filling a ticket for us. Some of them are going to be good interfacers with proper integration -- and yes we programmers are helping them to do so.
Side note: I don't think you are going to hear someone shouting that they are going to replace humans with AI. It started with this: people integrate AI into their workflow, layoff 10%, and see if AI helps to fill in the gap so they can freeze hire. Then they layoff 10% more.
And yes we programmers are helping the business to do that, with a proud and smile face.
Your argument depends on LLMs being able to handle the complexity that is currently the MBA -> dev interface. I suspect it won't really solve it, but its ability to facilitate and simplify that interface will be invaluable.
Im not convinced the people writing specs are capable of writing them well enough that an LLM can replace the human dev.
Error on line 5: specification can be interpreted too many
ways, can't define type from 'thing':
Remember to underline the thing that shows the error
~~~~~
| This 'thing' matches too many objects in the knowledge scope.
However, the AI is hard to work with, it expects specific wording in order to program our code as expected.
We have hired people with expertise in the specific language needed to transmit our specifications to the AI with more precision.