Abstract: [Purpose/Significance] This article aims to clarify the elements and process mechanisms in human–AI interaction, providing a reference for deepening Human-AI interaction theory and optimizing interaction design. [Design/Methodology] The study adopted the grounded theory approach and conducted a systematic analysis in-depth interview data from 20 users. By applying the Needs–Affordances–Features（NAF）framework to the human–AI interaction process, a theoretical model of the human-AI interaction process was constructed, with the core elements being interaction features, needs, agency, affordances and interaction behaviors between humans and GenAI. [Findings/Conclusion] The human-AI interaction process follows the overall logic of Needs-Affordances-Features（NAF）, in which user needs are driven by interaction features to form different types of GenAI affordances. Furthermore, this process can be refined into two stages: Needs-driven–Agency–Affordance Formation（NAGA）and Affordances–Behavior–Needs Satisfaction（ABN）. Among them, the agency of humans and GenAI is a key factor influencing the formation of affordances. Different combinations of affordances can produce three types of interaction behaviors: dependent, directive, and alternating, further fulfilling and influencing user needs. [Originality/Value] The study integrates agency into the NAF framework and proposes a mapping between affordance combinations and interaction behavior types, providing a new theoretical perspective for constructing more efficient human-AI interaction mechanisms and improving interaction experiences.

Keywords: Generative artificial intelligence, Needs-Affordances-Features framework, Human-AI interaction, Agency