In mainstream geometric modeling, cultivating virtual plant ecosystems is a difficult task. Algorithms for realistic scene generation are rooted in procedural models with no explicit or poor external control. We propose that virtual ecosystems modeling may be boosted using software agents as behavioral tools. An ecosystem grows and is driven by its internal rules of development. If it is left to its own fate, it will reach stability on the edge of chaos. Agents interact with ecosystems by adding plants, cutting or killing them, watering, stepping-over, or favoring some plant species. An agent is a characterization artifact that shows proactive conduct and is described by its set of sensors, effectors, internal states, and habits. Habits are defined as continuous functions and allow for characterizing a wide variety of behaviors.