Today, I am going to share with you my pet project that I built for fun with 4 AI agents talking to each other and challenging ideas, plans or whenever you want using the Walt Disney method for creativity. We have a kind of AI Agenting brainstorming room.
In simple terms, it’s an AI agent brainstorming room. Rather than relying on a single language model response, the system employs multiple AI roles that collaborate and debate from different perspectives. This makes the brainstorming process more dynamic, creative, practical, and fun to watch.
The main goal of this project was to experiment with LangChain, build a multi-agent workflow, and combine it with a lightweight web application using FastAPI, JavaScript, and Three.js. It was also a chance to do some vibe coding and explore how AI agents can be used for structured creativity.
What is the disney method?
The Walt Disney Method is a creativity and problem-solving technique that helps people look at one idea from different perspectives. It is often used for brainstorming, innovation, project planning, and creative thinking.
Traditionally, the Disney creativity method includes 3 main roles:
- Dreamer — generates bold, imaginative, and ambitious ideas
- Realist — turns promising ideas into practical plans
- Critic — identifies risks, weaknesses, and missing detail
For this project, I extended the original framework by adding a fourth role:
- Outsider — frames the challenge, clarifies the constraints, defines success criteria, and introduces fresh context
I added the Outsider role to introduce more variety and entropy into the conversation. In my version of the system, this role prevents the brainstorming process from becoming too limited too quickly.
Why Use AI Agents for the Disney Method?
One of the interesting things about the Disney Method is that it already feels naturally compatible with AI agents. Each role has a clear purpose, tone, and style of thinking. That makes it a good fit for a multi-agent AI architecture, where every agent is responsible for a specific perspective.
Instead of asking one AI model to do everything at once, this project separates the thinking process into stages:
- The Outsider frames the problem
- The Dreamer expands the idea space
- The Realist chooses a workable direction
- The Critic stress-tests the result
This approach can make brainstorming more structured,disciplined and more exciting and useful. It is especially interesting for: idea generation,startup brainstorming,creative writing prompts,product planning,feature ideation,early-stage concept validation and so on..
How the Multi-Agent Brainstorming Process Works
The multi-agent flow is designed to mimic a brainstorming session with different perspectives in the room.
Outsider start by shaping the problem and providing context. Then the Dreamer explores bold, creative, and unexpected directions. After that, the Realist narrows the ideas into something more actionable. The Critic evaluates the weaknesses, risks, and missing parts.
This cycle can help users move from vague inspiration to a more grounded and testable concept.
Project and technology
I used LangChain to manage LLM interactions and build AI agent workflows for the core intelligence of the app. The backend is powered by FastAPI, which handles the API layer and server-side agents. On the frontend, I used plain JavaScript for simplicity, plus Three.js to add an interactive arcade mode with 3D elements and a nice table interface.
Modes
Normal mode is general UI interface, where you can see board and agents talking to each other
Arcade mode is a kind of game mode with nicer UI and for more funny interaction.
How to run?
- Clone repo https://github.com/dmitrygalanov/disney-strategy-method-aiagents
- And follow instructions to install packages
- ./bin/python -m uvicorn disney_method.web:app –reload
- Configure your OpenAI keys or ollama access
- Put the idea and run the brainstorming
Normal Mode
http://127.0.0.1:8000/
Arcade Mode
http://127.0.0.1:8000/arcade
Github repository:
https://github.com/dmitrygalanov/disney-strategy-method-aiagents
Enjoy and have fun :)
