The Ambient Bot that we have developed makes it possible to access information by eye contact with virtual creatures in an augmented reality environment. By using Ambient Bot, information can be accessed without using complex operations from smartphones. In this paper, we focus on how that access of information consists of a pull-based interaction method and a push-based interaction method and show the feasibility of both interaction methods in Ambient Bot. In the pull-based interaction method that was adopted by the original Ambient Bot, multiple virtual creatures float around a user. Since each creature provides a different piece of information, it is possible for a user to easily access multiple types of content simply by making eye contact with each creature. On the other hand, in the push-based interaction method, a virtual creature predicts appropriate pieces of information that a user wants and actively notifies the user of that information. Our aim is mainly to investigate the design spaces for the push-based interaction method in Ambient Bot. In addition, we also investigate whether the push-based interaction method in Ambient Bot can be used to offer serendipitous information that will help people come up with new ideas when they receive diverse, interesting stimuli.