This research project was sparked by the emergence of generative visual tools based on machine learning in the late 2010s. How much creative autonomy did they afford, compared to prior computational generative devices and human creators? Can we now foresee the birth of an artificial artist, which would be completely self-ruled, independent of humans? Where would the poetics of such a creative entity lie? Informed by frameworks from systems theory and computational cognition, the investigation develops the concept of autopoietic aesthetic arrangement, which highlights a particular poetic mechanism within self-generating artworks. The vertiginous development of creative AI in the most recent years allowed this conceptual toolbox to be tested against new generative technologies, in particular the text-to-image semantic imaging tools.