I think the problem is that decision makers /people in power often don’t understand the technical jargon either. All that matters is, “is it working?” So once they see that they can get it “working” without paying for the expertise, they will take it.

The thing is also - the LLM will eventually get it wrong, but the frontier models are pretty good at fixing it if you can articulate the problem. Even non technically.

Under the hood? Impossible to understand. But it works!

Counterpoint - system architecture is really built around helping humans to build mental models of systems. If your plan is to have LLMs to all the work in the future, those mental models no longer matter the way they did in the past.

Probably, systems will break, and LLM costs will rise, and the arithmetic will shift - but that isn’t happening fast enough to stop the brain drain and negative pressure on employment and compensation.