Right now AI is too expensive or too risky to be used by businesses at a high scale. But costs will be driven down like anything in technology, and then the really powerful integrated systems will come out and many people won't be needed for supervision either. In fact, even now early adopters build tiered architectures where they (humans) are like the CEOs communicating directly only with a supervising agent (depending on how complex their architecture gets, it can be a sort of a CTO on top of team leaders who are in charge of worker agents or it can be directly a team leader for a less complex architecture).

Very soon we'll have OS's with fully integrated AIs, as in, workflows can be controlled by the OS's agent(s). That keeps people in the loop, unless the device is programmed to run fully automated, and we will see that as well.

Specialized AIs will likely be much more powerful than the current generalized (and even specialized) LLMs. We see that with Tesla and its automated driving system. I'm hearing quite impressive things about specialized AIs in the medical field.

So yeah, generally, people haven't been replaced yet as workers or supervisors, but it's likely we will see major shifts the following years.