Agentic Town — Hackathon Project Story

Inspiration

Agentic Town was inspired by a university-led research experiment where researchers simulated a digital town using autonomous agents to study how societies react when events are introduced into a controlled environment.

The idea was powerful: instead of predicting behavior with rules, let agents think, adapt, and respond naturally.
This project is my attempt to translate that research concept into an interactive, visual, and scalable demo.

---

Problem Statement

Most games and simulations rely on:

• Hardcoded scripts
  
• Predefined dialogue trees
  
• Predictable NPC behavior
  

These approaches fail to model real societal dynamics.

Agentic Town asks:

What happens when NPCs are replaced by autonomous AI agents with memory and decision-making ability?

---

Solution

Agentic Town is a living AI simulation where each NPC is an independent agent capable of reasoning and adapting based on context.

Each agent’s behavior is driven by:

[ \text{Action} = f(\text{Memory}, \text{Emotion}, \text{Environment}, \text{Time}) ]

This enables emergent behavior, where social patterns form naturally without explicit scripting.

---

How It Was Built

• Developed using Vibe Coding in Google AI Studio
• Gemini models used for agent reasoning and decision logic
• Multiple AI tools leveraged for rapid prototyping
• Phaser.js used to render the 2D world and agents
  

Notably, I had no prior experience with Phaser.js before this hackathon.
It was learned and applied specifically to bring the agentic world to life.

This project is currently a demo and work-in-progress.
Development depth is limited by free-tier Gemini API quotas, which restrict long-running multi-agent simulations.

---

Key Learnings

• Designing agent-based architectures instead of scripted systems
• Implementing memory-aware and context-aware AI behavior
• Rapid AI-driven development using modern tooling
• Translating research ideas into interactive systems
• Learning a new game framework under time constraints

---

Challenges

• Managing API rate limits and quota exhaustion
• Ensuring agents behave believably without rigid rules
• Balancing simulation complexity with performance
• Learning Phaser.js while building core logic simultaneously

---

Impact & Future Scope

Agentic Town demonstrates how agentic AI can be used for:

• Social behavior simulations
  
• Game NPC evolution
  
• Digital sociology experiments
  
• Training and scenario modeling
  

With higher API limits and more time, the system can scale to:

• Larger populations
  
• Long-term memory evolution
  
• Event-driven societal experiments
  

---

Conclusion

Agentic Town is not just a game demo—it’s a proof of concept for autonomous AI societies.

This hackathon submission showcases how research-inspired agentic AI can evolve into interactive, real-world simulations.

Built With

• gemini
• phaser.js
• typescript