Current Frontier AI companies are forecasting the day when there will be no other companies needed after them. Because their AI models and products will be the future of all business, managing every task and action, with humans orchestrating the tasks and validating the content.
The thing is, AI can never be at a point to do that.
The Snapshot Restriction
AI, no matter how refined it is or how well it performs, still needs to be trained on existing knowledge and data. That knowledge and data are a snapshot of the existing world to which they are exposed, with some historical context provided by previous snapshots.
The Cross-Discipline Inspiration Gap
A distinctive human inspirational thinking is done by looking at unrelated items and applying them to a problem. Think Newton and the falling apple, or Jiro Horikoshi looking at fish bones and mapping wing dimensions for the Japanese Zero fighter.
AI doesn't (yet) have that full breadth of knowledge to intersect from all possible disciplines.
The Sheer Volume of the Unknown
Pedro Domingos, in his book The Master Algorithm, puts it best: An all-knowing (omniscient) AI that can predict the future needs to have a full and perfect understanding of the entire universe. Everything. Until AI can do that, it will not have a perfect understanding of your problems and will therefore be fallible.
Why Bother with AI?
What we need is the capability to augment our own skills with just enough to expand our reach. AI can process connections faster than humans can, and purpose-trained AI is extremely valuable.
The best thing anyone can do in this current job market is to get to know AI, not for what it can do, but for what it can't. Use your Actual Intelligence to find the gaps that AI can't meet, and fill those gaps.