Teaching programming in early childhood (4-9) to enhance computational thinking has gained popularity in the recent movement of computer science for all. However, current practices ignore some fundamental issues resulting from young children's developmental readiness, such as the sustained capability to keyboarding, the decomposition of complex tasks to small tasks, the need for intuitive mapping from abstract programming to tangible outcomes, and the limited amount of screen time exposure. To address these issues in this paper, we present a novel methodology with an AI-powered integration platform to effectively teach computational thinking for young children. The system features a hybrid pedagogy that supports both the top-down and bottom-up approach for teaching computational thinking. Young children can describe their desired task in natural language, while the system can respond with an easy-to-understand program consisting of the right level of decomposed sub-tasks. A tangible robot can immediately execute the decomposed program and demonstrate the program's outcomes to young children. The system is equipped with an intelligent chatbot that can interact with young children through natural languages, and children can speak to the chatbot to complete all the needed programming tasks, while the chatbot orchestrates the execution of the program onto the robot. This would completely eliminates the need of keyboards for young children to program. By developing such a system, we aim to make the concept of computational thinking more accessible to young children, fostering a natural understanding of programming concepts without the need of explicit programming skills. Through the interactive experience provided by the robotic agent, our system seeks to engage children in an effective manner, contributing to the field of educational technology for early childhood computer science education.