It will get worse before it gets better.
We will see a gradual waning AI, with use cases that got added just to signal "innovation" to investors evaporating first, and all but the ones that actually make money (or break even) following soon after.
Available LLM tools will sharply contract.
I think we are going to see a rapid winnowing of "LLM in everything" once companies stop being able to subsidize the eye-watering costs of running these massive foundation models. Unfortunately it will create a crisis where LLM-dependent professionals cluster around the last few use cases.