One of the most impressive works of Escher depict mosaics of creatures coexisting in a shared space. The shape assumed by one creature is perfectly determined by the shapes of its neighbors, while it in turn helps shape its neighbors. What if we used a similar design to build a testbed for communities of world-optimizing agents? In an identification of map and territory, each agent would aim to remake the image-world in its own image, by representing itself across a larger and larger patch of the image-world. Imperfect sensors could be simulated by allowing read access to only a subset of neighboring pixels, and potentially adding noise. Similarly, imperfect actuators could be simulated by stunting write access (i.e. "agent #3 hasn't mastered this aspect of physics"). Could clever objective specs help avoid a monopoly on resources? What would be the kinetics of a sharp increase in granted optimization power?