Coding via LLM is not similar to using an abstraction. Imagine a car. The controls like steering wheel, the pedals, the gear levers. Those are abstractions.
But using LLMs are like driving using a remote control that has probabilistic behavior. You just loss what it feels to be in a car and you fail to improve as a driver because of the erratic remote control.