The cases I’ve heard of people getting actual value from LLMs involves a whole lot of deep thinking ahead of time, followed by lots of reflecting on what came out of the tool, so much so that it feels like tedious work and might just be the rubber-duck effect.

A number of “We tried AI, but the amount of error checking needed made it worthless” have been published recently.

If it feels like magic, it’s probably garbage. Anything worthwhile takes work.