From Player-Centered to Agent-Centered: A New Paradigm for Interactive Narrative in the Age of Large Language Models

This paper argues that large language model (LLM) driven interactive narrative is shifting the locus of pleasure from player-centred choice toward relationships with semi-autonomous agents that remember, infer, improvise, and sometimes fail. It contends that traditional interactive narratives position the player as the primary source of meaningful action, whereas systems such as 1001 Nights, Generative Agents, AI Town, Instantale, and SIMA 2 distribute narrative momentum across players, designers, agents, and computational infrastructures. To conceptualise this emerging form, the authors propose the Agent-Centered Narrative Pleasure (ACNP) framework, which links three dimensions: agency distribution, authorship configuration, and pleasure structures.

The analysis is exploratory and comparative, drawing on five recent agentic systems selected because each makes agent autonomy perceptible to players. Each case was coded across three layers: interaction grammar, agentic capacity (memory, planning, reflection, persistence, and opacity), and affective economy. The study does not report survey data or quantitative sample sizes; its evidence rests on qualitative case comparison and prior empirical literature on hallucination perception and player trust.

The case analysis identifies three configurations of pleasure. In 1001 Nights, pleasure arises from linguistic negotiation with an AI character whose coherence must be discovered, with hallucination treated as managed ambiguity. Generative Agents and AI Town shift pleasure toward witnessing emergent social narratives that accumulate without player intervention, while Instantale extends this into multi-timescale fantasy worlds. SIMA 2 represents collaborative delegation, where pleasure derives from anticipatory trust in an embodied agent's interpretation of high-level intentions.

The authors caution that agent-centred pleasures are intersectional and entangled with troubling dimensions such as opacity, labour extraction, and cultural bias. They conclude that agent-centred narrative complements rather than replaces player-centred design, and call for player reception studies with diverse participants and research on accountable agent authorship.