Yeah I don't think this get's enough attention. It still requires a technical person to use these things effectively. Building coherent systems that solve a business problem is an iterative process. I have a hard time seeing how an LLM could climb that mountain on it's own.

I don't think there's a way to solve the issue of: one-shotted apps will increasingly look more convincing, in the same way that the image generation looks more convincing. But when you peel back the curtain, that output isn't quite correct enough to deploy to production. You could try brute-force vibe iterating until it's exactly what you wanted, but that rarely works for anything that isn't a CRUD app.

Ask any of the image generators to build you a sprite sheet for a 2d character with multiple animation frames. I have never gotten one to do this successfully in one prompt. Sometimes the background will be the checkerboard png transparency layer. Except, the checkers aren't all one color (#000000, #ffffff), instead it's a million variations of off-white and off-black. The legs in walking frames are almost never correct, etc.

And even if they get close - as soon as you try to iterate on the first output, you enter a game of whack-a-mole. Okay we fixed the background but now the legs don't look right, let's fix those. Okay great legs are fixed but now the faces are different in every frame let's fix those. Oh no fixing the faces broke the legs again, Etc.

We are in a weird place where companies are shedding the engineers that know how to use these things. And some of those engineers will become solo-devs. As a solo-dev, funds won't be infinite. So it doesn't seem likely that they can jack up the prices on the consumer plans. But if companies keep firing developers, then who will actually steer the agents on the enterprise plans?