AI is squishy.

It's hard to use a squishy thing to interact with a hard thing.

For example, processes that must be extremely reliable and never fail are hard.

This is one of the reasons that unsupervised automation is tough with LLMs.

Even if you get it to work well 95% of the time, that last 5% of reliability is extremely hard to achieve.

If there's a human in the loop it doesn't have to be 100% reliable, the human can figure out when it works and bridge the gap.

If there's no human in the loop, there's no way to absorb the failure cases, and the viability of the whole system is reduced.