Generative AI currently employs linear chat interfaces that may bypass users' critical thinking. This design can result in automation bias, as individuals may accept AI-generated outputs without sufficient scrutiny. Consequently, users often become passive supervisors rather than active creators. This study addresses the challenge of human-AI co-thinking by shifting the emphasis from AI-generated results to collaborative cognitive processes. The project seeks to re-establish users as primary creators by designing a digital environment that reduces passive supervision. To evaluate human-AI co-thinking, a prototype named Flint was developed. Flint replaces the traditional chat stream with a dynamic spatial canvas, enabling users to externalize and manipulate their thoughts as discrete objects. As a mentor, Flint analyzes user needs and generates customized thinking tools presented as blank objects. Although the AI configures the environment according to user requirements, it cannot write on the canvas, requiring users to engage in genuine cognitive effort. This approach restores productive struggle and positions the user as the architect of the thought process. The findings indicate that transitioning from linear to spatial interfaces can restore human agency, positioning AI as a co-thinker rather than a replacement for human creativity.