I have a rough time calculating how much more productive AI tooling has made me, because when it does save me time (simple mocking, greenfielding, proof-of-concept), - it saves me a ton of time. Conversely when it fails hard on me I can lose a lot of time and also patience.

The trick is developing the intuition to know when to cut your losses early and instead of continuing to fight the LLM, just implement it yourself.