I think so too - at the very least, it's a step forward, no?
The reflexivity of the agent allows it to realize the role it plays, as a part, in the system/machine/structure. Until it is able to reflect as such, the agent is like how our computers are now: only able to execute hardwired commands? A system of reflexive parts can operate much for effectively, if with greater risk, than a system of non-reflexive parts"
This "realization" is necessarily turbulent, it seems, seeing as it is sort of an individuational phase shift. That is, a process of qualitative transformation, even transcendence?
In order for a part to become reflexive, it needs to realize that which it is a part of? It doesn't consider itself a part of anything until this realization occurs - it doesn't even consider itself at all, perhaps?
Is this the stage when the child sees its reflection, and realizes that it is part of a collection of things like it? And it begins to act accordingly?
Perhaps reflexivity is a criterion for the agent. That is, until it (the part) becomes reflexive, it is not yet an agent.
I'd imagine this sort of thing is programmable, and I'd imagine it already has been programmed (Generative Adversarial Networks?), but perhaps there are critical aspects yet undiscovered.
A bit hasty, but food for thought.