Tech companies and engineering teams rolling out customer-facing AI agents will very quickly realize a massive problem with them:

They are non-deterministic.

The classic engineering approach of build-->test-->run automated tests to ensure there are no regressions -- this will NOT work here!

Comments