Skip to content

2048-ish: A Quick Experiment in Agentic Gameplay

Published: Estimated Reading Time: 1 min read 1 min read

I had this idea to try building a game that would incorporate agentic gameplay into the familiar 2048 puzzle mechanic. The result is 2048-ish, a simple starting point for exploring how AI agents might play alongside or against humans. You can try the classic mode here.

The Basic Concept

The core game follows the standard 2048 rules - slide tiles, combine matching numbers, try to reach higher scores. What I wanted to explore was adding an agentic mode where AI agents could play the game, potentially learning strategies or competing with human players.

2048-ish game in action

Current State

Right now it’s just a basic implementation - nothing fancy, just enough to test the concept. The source code is available on GitHub if you want to take a look or build on it.

What’s Next?

This is really just a starting point. The agentic mode is still experimental, but it opens up interesting possibilities for AI-human collaboration or competition in puzzle games. Feel free to fork it and see where you can take the idea.

2048-ish menu showing Agent Mode

Check out the Agent Mode here to watch AI strategies in action!