AI driven development doesn't allow you fine enough control to really implement anything specific, you're stuck with whatever generalities the model derives from the prompt on top of whatever random changes and bugs it introduces between iterations. One could take care of those things outside of an LLM, but let's be realistic - for the vast majority of people "making the game fun" means adding "make the game fun" to the prompt and calling it a day. This is why the technical implementation by LLMs constantly improves but things like affordance, graphic design, level design etc so often fail.

So you would be right if you only get a single prompt of limited length, no assets, no documents, instructions to iterate until complete, and an hour of wall clock time.

So ...ok... there's a few fable 5.2 demos out there that do crazy things with even that.

But yeah... possibly we're more in agreement than not.

If you write hello world in python or slap some paint on a canvas or blow into a concert flute with no practice; it's painfully obvious that you're not done yet. You have a long way ahead of you to make something that looks or sounds like something.

In the case of LLMs and image generators, you often do immediately get something that looks like something. If you're a "majority of people" you'd be forgiven for thinking that you're done, and that that's all it can do.

But that's not true at all. If you're a "some people", you might just realize that You're not done yet . Not by a long shot. That's just a starting point. Bring in more tools. Work in layers. Write 10000 prompts on a single model and throw the results away, just to get a feel for the latent space. Do it again for 20 more models. Have an LLM do research into algorithms on the half that's less important to you. Have one dig up textbooks and papers on the half that's more important to you. Read them. Ask "Has anyone had this idea before?". Look up those artists and writers. Read their works. Be inspired. Now you might finally have a spec for your model that isn't just tropes. Get to work. Write your first 100000 words and put them aside. Now you might have enough skill to design a plot that holds together. Etc. etc.

And a 100 important things more that either I don't know or hasn't even been invented yet.

Work the problem step by step. ;-)

A new thing is: you can apply your artistic vision without needing a gigantic team to help you. That's very neat!

Of course I wonder, what if a large team did get together? How would they work? What could they build?

AI driven development allows as much control as you're willing to put in. You can have it make single digit changes if that's the amount of control you wanted