Either wrong or insufficient metric.

They could be converting on products the user didn’t even begin with an intention to buy, which would be impossible to compare with their existing flows.

This is a great summary of the current AI era.

Walmart's goal is to sell more things. Using AI did not help to achieve that goal. Is this a failure? No, we need to define a new metric based on what AI can do.

Same thing with software.

Are we shipping better software to happier customers? No? Better measure token usage, number of lines changed, and "developer velocity" instead.

I fear "discovery" will end up being a corporate euphemism for "shilling", worse than any search-engine manipulation since it has the potential to be so much subtler, pervasive, and parasocial.

The kind of microcosm parodied by The Truman Show is becoming plausible, at least digitally.