Sapling AI | Devpost

Sapling homepage
Sapling live game generation
Sapling games library
Sapling explore page

Inspiration

Game creation today is either technically complex or creatively limiting—especially for students, educators, and non-developers. We were inspired by the idea that imagination should be the only requirement to build a game. With the emergence of Gemini 3’s long-context reasoning and multimodal generation, we saw an opportunity to move beyond AI as a simple assistant and instead build AI as a full game developer—capable of planning, coding, validating, and iterating autonomously from a single natural-language prompt.

What it does

Sapling AI is an AI-powered platform that transforms plain-English game ideas into fully playable React Native games within minutes. Users describe a game concept, and Sapling AI automatically generates the complete game logic, UI, and visual assets. The platform also supports conversational iteration, allowing users to refine gameplay, visuals, and mechanics through follow-up prompts—without writing a single line of code.

How we built it

Sapling AI is built on Gemini 3 Pro, using a multi-agent architecture where each agent handles a specialized responsibility: request triage, naming, asset specification, game development, code inspection, and game editing. Imagen 3.0 is integrated to generate high-quality visual assets that are programmatically embedded directly into the generated game code. Agent coordination, parallel execution, and state management are managed using the Google Agent Development Kit (ADK), enabling fully autonomous, end-to-end game generation.

Challenges we ran into

The primary challenge was achieving reliability at scale—generating large, production-ready codebases while maintaining logical consistency across multiple agents. Synchronizing parallel code generation and asset creation without breaking gameplay flow required careful orchestration, validation, and error-recovery strategies.

Accomplishments that we're proud of

End-to-end text-to-playable game generation with no external tools
Fully autonomous multi-agent collaboration powered by Gemini 3
Seamless integration of AI-generated assets directly into live game code
Support for conversational, iterative game editing

What we learned

We learned that agent specialization significantly improves output quality, and that Gemini 3’s long context window is essential for generating complex, coherent applications in a single pass. We also gained valuable insight into designing AI systems that operate more like coordinated software teams than single monolithic models.

What's next for Sapling AI

Next, we plan to expand Sapling AI with multiplayer capabilities, physics-based engines, persistent game state (leaderboards and saved progress), and classroom-focused educational modes. We also aim to support additional deployment targets, including web and desktop platforms, making AI-generated games accessible everywhere.