Anyone else think that there is an AI adoption paradox?

Any use case with low enough stake for rapid and widespread adoption it is still much more capital efficient to use a low wage human labor over expensive AI systems. Every use case that is critical enough to warrant the costs of adopting it, there are still too many issues with reliability and edge cases for it to be worth using (the self-driving car last 10% problem.)

The only place where cost and reliability aren't an issue is the military (because it's worth it to try to prevent causalities) - a lot of weapons systems that have a higher failure rate than would be allowed in the private sector (think the Osprey, the F-35 and munitions failure rates)