GOh, I should have created the thread early on 
GI agree. At least some understanding of how programming works will be necessary for now, although there's always the chance that changing -> might change everything over night.
GI believe the most tangible benefit would be to empower people to use languages that they don't know yet. I may know exactly what I want to do with Python, but before I can do anything useful I'll have to spend a few days learning the syntax and how the libs work.
GSo just to summarize the idea, it would go like this:
- I write the pseudo code that does what I want
- I write some more pseudo code for unit tests
- Windmill analyzes what libs I'm using in my pseudo code and fetches useful examples from the official docs
- Windmill transforms my pseudo code into actual code and runs the unit tests
- If the unit test fails, feed the error into the AI and ask it to fix
GAnd if Windmill ever does actually take that direction, perhaps it could even expand to be focused not only on internal tools, but software development in general. I know at this point I'm looking faaar into the future, but maybe it could be basically a new way to write software. 
GThat might seem crazy, but Bubble.io is making a lot of money with that mission (but I think their 100% no-code approach limits it too much).
GYes, what I mean is that if you make it work for 80% of the cases with GPT-4, maybe next year it will just require changing a parameter in the API request to make it work with 98%
GAnd I think betting on OpenAI at this point seems very safe
GI created a topic in the forum because I believe it's better for long-lived conversation: helpAI-based programming
GCheck the second link that I posted there (Respell)