What if an agent mental space does constitute the environment for a second agent? Even the simplest agent architectures are composites of entities that, together, define the agent?s behavior. Now, suppose that one can use a second agent to percept and manipulate those entities. Why should one want to do so? Are there any advantages? And what are the disadvantages?

This paper reports our initial investigation into the advantages of such scenario. We conducted a simple experiment based on Steels? Mars Explorer subsumption agent: His list of reaction-pairs are initially shuffled and a metaagent can promote or demote those pairs in response to individual-power local variations. After a while the observed list of reaction-pairs and the implied agent behavior are very close to the original, and intended, design.