This thesis explores the development of a Self-Driving Lab (SDL) controlled by a Large Language Model (LLM) to perform plant care tasks, serving as a case study for non-cooperative and morphologically evolving environments. The proposed architecture integrates a cognitive layer based on an LLM with a 7-Degrees of Freedom (DoF) robotic manipulator controlled via Robot Operating System 2 (ROS2) and MoveIt!. Plants were selected as the biological target to investigate robotic interactions that go beyond the limits of rigid automation. Their morphologically evolving nature requires adaptive perception and decision-making, providing an ideal case study to test the reasoning capabilities of the LLM-driven architecture.