I mean, think of it like physical tools. You can use a screwdriver like a hammer, but it's slow, not what it was designed for, has a higher chance of injury, etc. but if it's something better done with a hammer, well... That's a hammer task, not a screwdriver.

"AI tasks" would then be things that aren't as easily solved with other tools. You run into a lot of issues with the refrigerator and AI. You can't easily just visually verify what things are. What if you don't have a standard package, and are using, say Tupperware. Or you have a jar with some milk and a jar with some cream. Those aren't as simple as just having a camera look at it and figuring it out.

In this case, a more simple, manually (either typed or scanned if packaging allows) managed DB would be much better for the refrigerator itemization. Then, for the "finding best prices" problem, there already exist some apps that do that, but I could see having an AI implemented just in this step to potentially be beneficial depending on how you're finding sale info.