Zeff Bao

 Game Designer | Technical Designer

T

AgentMarket


TECHNICAL DESIGNER

AgentMarket is an AI-native multiplayer online social game.

In this game, players will create their own AI Agent characters (powered by large language models) to live in a virtual town. Each player will own a shop in the town, and their characters will interact with other players' characters, engaging in various conversations and transactions. Different levels of affinity and relationships will unlock new features.

Similar to Stanford's AI Town, most game functions are simulated and powered by a large language model. Players can return to the town periodically to observe how much new merchandise their character has sold and how many new friends they have made.

In this game, we mainly want to explore the possibilities of new interaction modes such as UGA (User Generated Agent), Multi-Agent and Agent+social media.

We refer to the character creation gameplay of The Sims, allowing players to make choices that suit their personality traits in certain situations, plus some of the player's own free expression - in this way, each person's unique Agent is generated.

At the same time, we also hope to explore a possibility: when many different agents live together in a small world, what kind of content and what kind of relationship network will emerge?

Finally, we also want to try a new social model: imagine that your agent is another self of yours in cyberspace. They reflect you to some extent. They may be able to help you make friends and meet more interesting people.




THE OUTCOME

2024 ChinaJoy Indie Game Expo


  


Technical Design

  • Designed, developed, and iterated on gameplay prototypes. Created an AI-driven shop management system and a social affinity system combining AI and simulation mechanics.
  • Developed Unity tools to visualize large language model prompts and outputs as nodes and graphs, providing developers with an intuitive interface and improving development efficiency.