Unless AI companies are willing to do the impossible and accept liability for incorrect responses/outcomes, they're never going to be trusted with any task that could incur liability (law, accounting, medicine, architecture, engineering).
Could humans review AI output? Sure. You still need humans. Funnily enough wages won't decline in the long-run - they'll rise as firms compete for those who remained employed in said roles as others fell out of favour and have gaps in work experience.
Many, many variables at play. I never take anyone seriously who never considers this (many people on here).
If that doesn't happen, then companies will just continue to make excuses for why they need to lay off 20,000 workers after having their best quarter ever.