What I'm seeing, as someone who works in a giant company but not in a tech department, is the rise of middle companies that specifically do all the hard work of organizing/wrangling the AI and providing more streamlined options that are easier to understand, with the end users giving up flexibility and freedom for ease of use. (Like the AI version of Dreamweaver, I guess would be a comparison?) Or modern no/low-code offerings.
I've also seen already existing SaaS offerings integrating AI into their already existing product. How well that's going to work... I'm not sure. The economics of at least some of what I've seen is questionable (e.g. giving end users essentially unlimited token spend without passing any of the cost along), and at least one company in my field is clearly adding AI features because they're spooked that the existence of AI is damaging their moat/vendor lock in.
The shift might look like 'company Y contracts with AI companies A, B, and C, and as a result does more with less and therefore needs 30% fewer people'. In theory, companies A, B, and C are creating jobs so the job numbers shift, but your point about a certain level of understanding being needed to do those new jobs definitely stands.
The question is are there enough competent people to staff the middleman companies (I'd say yes specifically because entry level tech is a bloodbath atm) and how simple can these tools be made? Can they be dumbed down enough so that businesses can just buy a package from a vendor where they can give tool access to the people who can't program VCRs and trust that they won't shoot themselves in the foot? That, I don't know.