Since the realisation of the computer, and shortly after the inception of artificial intelligence (AI), there has been an explosion of research solving human-level tasks using autonomous entities that are able to learn about an environment by observing and influencing it, known as intelligent agents (IA). This potent AI technique has yet to filter into the field of thermoscience, where the conceptual design and optimisation of complex energy systems has been a particularly challenging problem. Much of the design process still requires human expertise. But with the continual increase in computational power and the use of IAs, it is now time to shift the responsibility from the human to the computer. This research attempts to answer the question of whether it is possible for a computer to conceptually design a complex energy system autonomously, from inception. The complex energy system to be designed and optimised is a thermoacoustic heat engine (TAHE), which converts thermal to acoustic power. The complexity of its physical behaviour and its many design parameters makes it a challenging energy system for conceptual design and optimisation and consequently an ideal candidate for this particular research. The TAHE is designed for low temperature waste heat utilisation from a baking process. In this work an approach is employed that is based on a reinforcement learning intelligent agent (RLIA). The RLIA is first employed to simultaneously optimise thirteen design parameter values. The RLIA was able to learn key design features of a TAHE which lead to the reduction in acoustic losses and an acoustic power from the engine of 495.32 W, when the thermal power input was 19 kW. For the main experiment, the RLIA must conceptually design the TAHE from scratch, changing both the parameter values and the configuration of the device. The results have shown the remarkable ability of the RLIA to identify several key design features of the TAHE: the correct configuration of the device, selecting designs that reduce acoustic losses, create positive acoustic power in the stack region and determine the region of optimality of the design parameter values. The RLIA has shown a great capacity to learn, even when contending with a complex environment and a vast search space. With this work we have introduced RLIAs as a new way approach to such multidimensional problems in the field of thermoscience/thermal engineering.