I don't think lack of a killer app is the problem here, the problem is AI isn't, and probably won't be for some time, a one size fits all plug and go solution to anything. It still takes a reasonable amount of time and effort to get AI setup in any meaningful way.
I think we need education etc to catch up a bit first before AI becomes a widely deployed thing.
We've seen it before in tech loads of times, notably in web development...there was a golden era where relatively few web developers existed because being able to develop a website meant you needed to have pretty deep knowledge...then education caught up and we had the JS Dev explosion which lead to stuff like funny moustaches and Wix...now every man and his dog can build a website in under an hour.
Something similar will happen with AI...for now, it's still fairly obscure technology and requires effort and knowledge...a few years down the line the soy latte waxed moustache gang will turn up, then eventually every man + dog will be able to roll their own AI.
The time to make money with it is now, the time to start looking at the next technical developments is when your barber has a diffusion model with inpainting built into his mirror to try hair cuts before you commit to them and a large language model to keep you occupied with bullshit small talk while he works...we're kind of there now with web development...most devs think they're working on a multi billion dollar project that will launch them to Facebook infamy...but really they're making menu apps for Indian Takeaways and pubs using prefabricated objects they import with npm through shadcn. We're long past the point of designing things from the ground up, it's all frameworks, modules and extensions. At some point the gatrillions of AI models will start to converge a bit and we'll end up with just a handful of them that people will build frameworks, modules and extensions around to make deploying customised AI a paint by numbers affair.