Agent Forge

An AI product council where specialized agents collaborate to transform ideas into complete MVP plans.

Comment

Inspiration

Building a great product is rarely the result of a single person's decision. It requires collaboration between product managers, engineers, designers, QA specialists, and technical leaders, each bringing a different perspective.

Most AI tools generate a single response from one model. We wanted to explore a different approach: what if AI could simulate an entire product team?

Agent Forge was created to answer that question by orchestrating multiple specialized AI agents that collaborate, debate, vote, and ultimately transform an idea into a complete MVP implementation plan.

What it does

Agent Forge is a multi-agent AI Product Council that transforms a product idea into a structured MVP plan.

After receiving a product idea, six specialized AI agents collaborate:

  • CEO – final decision maker
  • Product Manager – product vision and requirements
  • CTO – technical architecture
  • Designer – UX and user experience
  • QA – quality assurance and risk analysis
  • Engineer – implementation planning

The agents discuss every agenda item, identify conflicts, vote on competing solutions, and generate a complete implementation package including:

  • Product Summary
  • MVP Scope
  • User Flow
  • Technical Architecture
  • Data Model
  • API Design
  • Implementation Roadmap
  • Risk Analysis
  • Decision Log

How we built it

The frontend was built with Next.js, React, TypeScript, and Tailwind CSS.

The backend uses Node.js and Express, while Server-Sent Events (SSE) provide real-time streaming of the AI council discussion.

All AI reasoning is powered by Qwen models through Alibaba Cloud DashScope's OpenAI-compatible API. Each AI council member is driven by dedicated prompts and coordinated through a structured backend workflow.

The backend is deployed on Alibaba Cloud ECS, while the frontend is hosted on GitHub Pages.

Challenges we ran into

Designing independent AI agents was much more difficult than simply writing prompts. Each role needed to maintain its own expertise while contributing to a coherent discussion.

Another challenge was orchestrating the complete workflow—from agenda generation and multi-agent discussions to conflict detection, voting, CEO arbitration, and final document generation.

Deploying the backend on Alibaba Cloud ECS and configuring the application for public access was also an important part of the project.

Accomplishments that we're proud of

We're proud of building a working multi-agent system that behaves like a real product team instead of producing a single AI response.

Highlights include:

  • Real-time streamed AI discussions
  • Specialized role-based reasoning
  • Conflict detection and voting
  • CEO arbitration for final decisions
  • Automatic generation of complete MVP documentation
  • Successful deployment on Alibaba Cloud using Qwen models

What we learned

Building multi-agent systems is less about creating better prompts and more about designing effective collaboration between AI agents.

We also gained valuable experience deploying production applications on Alibaba Cloud ECS and integrating Qwen models through DashScope's OpenAI-compatible API.

The project demonstrated that specialized AI agents can generate more structured and realistic product planning than a single AI model working alone.

What's next for Agent Forge

We want to evolve Agent Forge into a complete AI-powered product team.

Our roadmap includes:

  • Persistent project memory
  • Interactive collaboration with human stakeholders
  • Additional specialist AI agents
  • Architecture and UI diagram generation
  • Jira and GitHub integration
  • Automatic task creation from implementation plans
  • Support for iterative product refinement across multiple sessions

Built With

Share this project:

Updates