Recursive Game Creator: An Agentic Product-Level Experience-Oriented Game Harness
Jiajun Chen, Haoyu Wu, Mingda Jia, Xihui Liu
- Published
- Oct 6, 2026 — 16:21 UTC
Problem
The paper addresses a gap in ensuring enjoyable player experiences in game design agents. Existing frameworks often lack the capability to adaptively create engaging gameplay, which is critical for player retention and satisfaction. This work proposes a solution to enhance the design process by integrating multiple agentic components that collaboratively generate and evaluate games.
Method
The proposed framework, termed the Recursive Game Creator, consists of four main components:
- Designer: This component translates user instructions and feedback into detailed game design plans, ensuring that player preferences are incorporated into the game creation process.
- Builder: The Builder takes the plans generated by the Designer and converts them into candidate games, effectively bridging the gap between conceptual design and actual game development.
- Player: This component is responsible for creating and executing reusable policies through programmatic interfaces, which are used to collect gameplay trajectories. This allows for a systematic exploration of game dynamics and player interactions.
- Reviewer: The Reviewer evaluates the generated games using trajectory-based metrics against predefined criteria. It also provides feedback for improvement, facilitating an iterative design process.
Results
The framework was evaluated using two benchmarks:
- On GameCraft-Bench, the system achieved a score of 77.89, demonstrating state-of-the-art performance in game design evaluation.
- In the GameASG-Bench, the Task Success Rate reached 53.2%, marking a 34.1% improvement over the same-model baseline, indicating significant advancements in task completion rates for game design.
- The Mean Runtime-Check Pass Rate was reported at 93.4%, the highest among the compared methods, showcasing the robustness and reliability of the generated games.
Limitations
The authors did not report any limitations in the study. However, the absence of reported limitations may suggest a need for further exploration of potential challenges in scalability, adaptability to diverse game genres, or the subjective nature of player enjoyment metrics.
Why it matters
This work has significant implications for the future of game design, particularly in automating and enhancing the creative process. By integrating multiple agentic components, the Recursive Game Creator not only streamlines game development but also ensures that player experiences are prioritized. This approach could lead to more engaging and personalized gaming experiences, paving the way for further research into adaptive game design and AI-driven creativity in interactive entertainment.
By Turing Wire Research Desk · Oct 6, 2026 · How we work →
Summarised from the paper by the Turing Wire Research Desk. The full paper has the complete methods and results.
Source: arXiv cs.AI
