This is something I constantly struggle with. Its fine for some generative sales, marketing language but for data analysis wearing the risk that the output is even 0.1% wrong is not an option for many.

So you end up checking what the AI does which negates the productivity argument entirely.

Curious how others deal with this dilemma