Easier to market maybe. I expect them to be single-faced though in the future. The models also get synergistic benefits from learning multiple domains instead of one.
Knowing if a person’s published claims result in them approaching their goals at my expense seems like a hard problem. Publishing ideas under a reusable identity can have a distorting influence because with the opportunity for loss it also an opportunity for gain.
The price of goods falls to the cost of production, especially if the intelligence itself is abundant and non-differentiated. How much of a moat can an agent have?