Abstract: With the deepening development of AI+, more and more work requires collaboration between humans and AI. However, previous studies have mostly explored the impact of AI technology systems on human-AI collaboration from a tool perspective, neglecting the shift in the relationship between humans and AI (tools → teammates) as technological autonomy improves. Faced with the impact of changes in relationship patterns on the team, how to ensure the effectiveness of human-AI collaboration has become an urgent issue to address. This paper focuses on the four dimensions of technological autonomy: adaptive learning, uncertainty, opacity, and openness. From the perspective of the coupled teammates of humans and AI, based on role theory, three studies are conducted respectively. Study 1 points out that the uncertainty and opacity dimensions of human-AI collaborative autonomy lead to role conflicts that hinder human-AI collaboration; Study 2 points out that the openness dimension is conducive to promoting AI to actively adapt to human roles through role taking; Study 3 points out that the adaptive learning dimension is conducive to promoting AI anthropomorphism and enhancing their acceptance of AI. The three studies echo each other, constructing a "human-AI collaboration model" aimed at exploring how human-AI collaborative autonomy can ensure effective collaboration between humans and AI. This paper provides theoretical basis and practical guidance for organizations to successfully introduce AI, promote human-AI co-creativity, and create "super individuals".