1. Cleaning up the junk left after the AI bubble burst. 2. Debugging code you didn’t write and didn’t really spec. 3. “Would you like fries with that?”

Somewhat tongue-in-cheek, there will be three or four development tiers left: 1. model development and optimization, 2. data pool management, 3. downstream consumers of item 1, and 4. everything else