If you'd like the point of view of a complete idiot (I'm neither a coder nor a data scientist):

Technically we can overcome this by reviewing AI’s code. But what happens when humans can’t read code anymore?

Without “software engineers”, all technology is a black box

If no one codes anymore, can we understand code written by AI?

With AI writing software, we won’t put in the thousands of hours required to be good at it. If we’re not good at it, how can we review it.

In this scenario of lost knowledge, all technology essentially becomes a black box.

In theory, even if in the future we no longer have people who are trained to write code, we could create a class of human drones who are trained to read code (just as today there are people who read Sanskrit or Homeric Greek, even though no one speaks or writes either of those languages with "native" fluency, or like the proofreaders at a newspaper who are not necessarily qualified to write copy, or literary critics who can’t write a novel).

Again from the idiot's point of view, there might be a program or application that could translate the response of these human readers into a format that could be "understood" by and applied to an AI system. But how would we (humans) be able to determine whether or not such a kludge was viable?

Are we comfortable getting to this point as a society?

A society can be lulled into comfort by any agenda or system of thought, regardless of how harmful it is to the society (as Huxley showed in "Brave New World," or Zamyatin in "We"). Ideally, there will be enough rebels to keep the comort of dystopia from becoming the norm.