Assistive high-technology has demonstrated significant efficacy in improving the quality of life for individuals with autism spectrum disorder. Unlike conventional low- to medium-tech systems, assistive robots allow personalized interaction, promote attention and directed learning, serve as intermediaries between the child and their social environment, and monitor results through data collection and task tracking. Correct design needs a set of requirements linked to the needs and singularities of autism (language, autonomy-human control balance, response times, safety, and comfort, among others. In this context, emerging trends are integrating artificial intelligence to improve the functionality of social robots. This paperwork performs an analysis to identify the fundamental requirements in the interactive and intelligent design of assistive robots: agent-based interaction, adaptability, progress, and personality together with the benefits provided. In addition, the main negative impacts derived from an incorrect integration of artificial intelligence in assistive robots are pointed out, such as the loss of human interaction, the risk of dependency, and the costs of implementation and maintenance.