How I Used Switch to Build AgriCommand AI: Five AI Agents Collaborating for Smarter Agriculture

What if AI agents could do more than answer individual questions?

What if they could work together, exchange specialized knowledge, coordinate tasks, and help humans make better decisions?

That question inspired me to build AgriCommand AI – A Human-AI Collaborative Agriculture Intelligence Platform for the Switch Hackathon.

In this article, I'll explain the problem, our solution, the five specialized AI agents, how we integrated Switch, the website features, and the technologies behind the project.

🌱 1. The Problem: Agricultural Decisions Are Connected

Agriculture is not a single-problem domain.

A farmer making an agricultural decision may need to consider several interconnected factors:

• 🌾 Which crop is suitable for the soil?
• 💧 Is sufficient water available?
• 🌦️ What weather conditions and risks should be considered?
• 📈 What are the current agricultural market trends?
• 📋 How can different information be converted into a practical action plan?

These decisions require different types of knowledge.

A single general-purpose AI assistant may provide information, but coordinating multiple specialized responsibilities within one workflow presents a different challenge.

We wanted to explore a different approach:

A team of specialized AI agents working together with a human coordinator.

💡 2. Introducing AgriCommand AI

AgriCommand AI is a collaborative agricultural intelligence platform designed to support farmers and agricultural coordinators.

Instead of depending on one AI assistant for every task, our platform uses five specialized Gemini-powered AI agents.

Each agent has a specific responsibility and contributes to a shared agricultural decision-support workflow.

Our objective is to bring agricultural information, AI analysis, agent collaboration, and human decision-making into one platform.

🤖 3. Meet Our Five Specialized AI Agents

We designed five agents with different responsibilities.

1. Agri-Triage – The Workflow Coordinator

Agri-Triage is responsible for understanding incoming farmer requests.

Its responsibilities include:

• Understanding farmer-submitted agricultural problems.
• Classifying the reported issue.
• Identifying priorities.
• Coordinating the workflow between specialized agents.

2. CropMind – Crop Intelligence Agent

CropMind focuses on crop suitability analysis.

It considers:

• Soil conditions.
• Water availability.
• Seasonal suitability.
• Previous cropping information.
• Suitable crop options.

3. MarketPulse – Market Intelligence Agent

MarketPulse focuses on agricultural market information.

Its responsibilities include:

• Analyzing mandi price information.
• Organizing agricultural market insights.
• Helping farmers understand market trends.
• Supporting informed selling decisions.

4. WeatherGuard – Weather Intelligence Agent

WeatherGuard focuses on weather-related agricultural risks.

It is designed to:

• Evaluate weather conditions.
• Identify potential agricultural risks.
• Provide information that supports agricultural planning.

5. AgriReport – Reporting and Synthesis Agent

AgriReport brings the findings together.

Its responsibilities include:

• Combining findings from specialized agents.
• Organizing agricultural analysis.
• Generating a consolidated agricultural action plan.
• Presenting information for human review.

Each agent contributes a different perspective to the same agricultural problem.

🔗 4. How We Used Switch

This is one of the most important parts of our project.

We used Switch as the collaboration layer for our custom Gemini-powered agricultural agents.

Through a custom Switch adapter, our application connects to a live Switch room.

Our goal was to explore how specialized AI agents could collaborate rather than operate as five independent assistants.

Our collaborative workflow

The intended workflow follows these steps:

1. A farmer's agricultural concern enters the AgriCommand AI platform.
2. Agri-Triage understands the request and coordinates the workflow.
3. CropMind, MarketPulse, and WeatherGuard contribute their specialized analysis.
4. AgriReport combines the findings into an agricultural action plan.
5. A human agricultural coordinator reviews important recommendations.
6. The final decision remains with the human coordinator.

This approach explores how agent collaboration can support real-world workflows where human judgment matters.

Integration status: Our current implementation uses a live Switch room. We have not yet connected external messaging platforms such as Slack, Discord, Telegram, or Microsoft Teams.

🖥️ 5. Exploring the AgriCommand AI Website

We also developed a centralized web application to make the agricultural workflow accessible and understandable.

The platform includes several dedicated workspaces.

Workspace	Purpose
Overview Dashboard	Centralized view of agricultural activities and insights
Farmers	Organizes farmer profiles and information
Agricultural Reports	Manages farmer-submitted problems
AI Agent Workspace	Displays specialized agents and their activities
Crop Intelligence	Presents crop suitability analysis
Market Intelligence	Organizes agricultural market insights
Weather Intelligence	Displays weather-related agricultural risks
Collaboration Rooms	Provides a shared agent collaboration environment
Human Approvals	Allows coordinators to review recommendations
Analytics and Reports	Organizes agricultural insights and reports

Our goal is to bring the entire decision-support workflow into one platform.

🛠️ 6. Technology Stack

We combined modern web technologies with AI-powered agent collaboration.

Frontend

• React 19
• TypeScript
• Vite
• Tailwind CSS
• Lucide Icons
• Recharts

Backend

• Node.js
• Express.js
• TypeScript

Artificial Intelligence

• Google Gemini API
• Five specialized AI agent roles
• Multi-agent workflow orchestration

Agent Collaboration

• Switch
• Custom Switch Adapter
• Live Switch Room

Database Architecture

• Supabase
• PostgreSQL

Supabase and PostgreSQL are the target database architecture for our migration.

Development and Testing

• Git
• GitHub
• Postman

🧠 7. What We Learned from Building with Switch

One of our biggest takeaways was understanding the difference between building a traditional AI chatbot and building a collaborative AI workflow.

A traditional chatbot primarily focuses on responding to a user's request.

A collaborative agent system introduces additional responsibilities:

• Assigning tasks to specialized agents.
• Coordinating their contributions.
• Sharing information across a workflow.
• Combining different perspectives.
• Keeping humans involved in important decisions.

Switch helped us explore this collaboration model and think beyond isolated AI interactions.

We also learned that designing a multi-agent system requires clearly defined agent responsibilities and a well-organized workflow.

🌍 8. Why AgriCommand AI Matters

We believe AI should not simply generate answers.

It should help organize information, coordinate specialized capabilities, and support people in making better-informed decisions.

Agriculture is one area where this approach has meaningful potential.

Our vision is to make agricultural decision support more organized, accessible, and collaborative.

🚀 9. What's Next?

We want to continue improving AgriCommand AI by exploring:

• More reliable agricultural data integration.
• Improved agent communication and coordination.
• Expanded agricultural intelligence.
• Stronger human approval workflows.
• External messaging platform integrations.
• More comprehensive reporting and analytics.

These are future development goals, not claims about features already completed.

🔗 10. Explore the Project and Resources

Project resources

• 💻 GitHub Repository: [https://github.com/asusvivobook176-svg/AgriCommand_AI]

Technologies and documentation

• Switch Documentation
• Switch GitHub Repository
• Google Gemini API Documentation
• Supabase Documentation

🙌 Final Thoughts

Building AgriCommand AI gave us an opportunity to explore a different way of developing AI-powered applications.

Instead of building one assistant that attempts to do everything, we explored how multiple specialized agents can contribute to a coordinated workflow while humans remain responsible for important decisions.

Our vision is simple:

One farmer's problem.

Five specialized AI agents.

One coordinated workflow.

Human decisions at the center.

Thank you to the Switch team for providing an opportunity to experiment with collaborative AI systems and build something meaningful.

I'm excited to continue learning, building, and exploring the possibilities of multi-agent AI.

Built for the Switch Hackathon.