The concept of Quantified Self is about connected objects self-monitoring their human owner (e.g., a watch measuring heart rate, etc.). A natural transposition is in self-monitoring arbitrary things, therefore named Quantified Things. In this paper, we present the case of self-monitoring agricultural products. We discuss the rationales for the design of a Quantified Fruit multi-agent architecture for self-monitoring and self-prediction of the maturation of fruits. The architecture includes 6 different types of agents, the 2 more specific ones being respectively, the self-controller equipped with various sensors and the self-prediction module. Our current implementation uses an Arduino microcontroller board with 5 sensors (measuring respectively: temperature, light, humidity, hydrogen and methane). The prediction module uses a neural network. We have implemented the architecture and have conducted various experiments, storing bananas in diverse settings: room, refrigerator, in a box, with other fruits, etc. The paper discusses the architecture, its current implementation, experiments and current results. Future issues (scalability, collaborative prediction, etc.) are also addressed.