Building a Scalable, Multi-Agent AI System with the A2A Protocol

Executive Summary

To solve the critical industry challenge of scaling AI systems, a RUNNINGHILL developer dedicated 3-4 weeks to a strategic R&D project. The initiative involved building a proof-of-concept for the new Agent-to-Agent (A2A) protocol. The result was a successful multi-agent system where agents built on different frameworks (Google ADK and CrewAI) collaborated seamlessly to perform a complex task. This demonstrates our expertise in building modular, future-proof, “microservice-style” AI architectures that go beyond the limits of monolithic agents.

The Internal Challenge

Monolithic AI agents are powerful but fundamentally limited. As we found, they are difficult to manage, inefficient to diversify, and, most importantly, scaling them often results in a breaking change.

Even standard multi-agent systems don’t solve the core issue. They are often “closely coupled,” meaning their agents are co-dependent. This rigid structure makes it a time-consuming development task to add new services, remove old functionality to cut costs, or offer a single, isolated service to a client. We needed a way to build AI systems with the same flexibility and scalability as a modern microservice architecture.

The Solution: Innovation in Action

We built a proof-of-concept system using the brand new Agent-to-Agent (A2A) protocol, an open-source standard released in April 2025 by Google, Microsoft Azure, and Amazon AWS to solve this exact problem.

Our solution is a multi-agent system that schedules a tennis match. It consists of four independent agents:

• A “Host Agent” (the orchestrator)
• Two “Personal Agents” (to check calendars)
• One “Tennis Court Agent” (to handle the booking)

The system works by using the three core concepts of the A2A protocol:

1. Agent Card: Each remote agent publishes an “Agent Card”—a machine-readable “business card” that describes its unique skills and how to communicate with it.
2. Standardized Messages: The Host Agent discovers these agents and communicates with all of them using a universal JSON RPC message format, regardless of what framework they’re built on.
3. Agent Executor: A special “wrapper” on each agent translates the standard A2A message into the specific format its native framework (like Google ADK or CrewAI) can understand.

This architecture allows agents built with different technologies (Google ADK and CrewAI) to collaborate seamlessly to book the tennis match, with the final booking confirmed in a MongoDB.

The Build

This project was a prime example of our commitment to proactive upskilling. It began as a grassroots R&D initiative by one of our developers, Rhulani Hlungwani, to master and test the new A2A protocol just months after its public release.

In just 3-4 weeks of R&D, he went from concept to a fully functional, multi-framework POC. The process involved deep research into the new protocol, architecting the data flow, and building and deploying four separate agents on two different platforms to prove the promise of true interoperability.

The Results & Impact

This project successfully demonstrated a more advanced, scalable, and flexible future for AI systems.

“That was a really nice demo. A lot of work that went in here, but hopefully you’ve learned quite a bit.”

– Henk Bothma, Head of IT

Bulleted Key Metrics:

• 100% Scalable Architecture: Proved a “microservice-style” architecture for AI is possible. New agents (services) can be added or removed on demand without re-engineering the entire system.
• True Interoperability: We successfully made agents from completely different frameworks (Google ADK and CrewAI) collaborate to complete a single, complex task.
• Rapid Upskilling & Innovation: Our team demonstrated the agility to learn, implement, and validate a brand-new, cutting-edge protocol within 3-4 weeks of effort.
• Powerful “Proof of Work”: We now have a proven solution and the in-house expertise to solve the critical industry problem of scaling and managing complex AI agent systems.

Why This Matters to Our Clients

This R&D project directly translates to client value. Any company building with AI will eventually face the same problem of scale. Our success proves:

• We Build Future-Proof AI: We don’t just build monolithic agents that will become technical debt. We build scalable, modular, microservice-based AI systems that can grow and adapt with your business.
• We Are Expert Integrators: Our expertise with the A2A protocol shows we can make different systems—even those from different vendors or built on different frameworks—communicate and work together as a single, cohesive application.
• We Are Proactive Experts: You don’t have to wait for a new technology to become mainstream. We are already mastering the next-generation tools (like the A2A protocol) to ensure the solutions we build for you are on the cutting edge.

About RUNNINGHILL

Established in 2013, Runninghill Software Development is among South Africa’s fastest-growing BBBEE Level 2 companies within the Information Technology Sector. With over 50 full-stack developers specialising in finance, banking, fintech, and more, we help companies design, build, and ship amazing products.

We are passionate software developers who deliver world-class solutions using the best tools in the industry. We focus on technical team augmentation, consulting, and agentic solution design, and we partner with the best design agencies so we can focus on what we do best: technical excellence.

Our services include:

• Technical Consultation: We partner with your business to identify opportunities, solve complex challenges, and design strategic technology roadmaps that drive growth and digital transformation.
• AI-enabled Technical Team Augmentation: We provide skilled developers and technical experts who integrate seamlessly with your existing team to accelerate project delivery and fill capability gaps.
• Agents as a Service (AaaS): We design, build, and deploy intelligent AI agents that work autonomously within your business processes, creating a digital workforce that handles complex tasks while you focus on strategy.

Frequently Asked Questions (FAQ)

1. What is the Agent-to-Agent (A2A) protocol? The Agent-to-Agent (A2A) protocol is a new, open-source, standardized way for AI agents to communicate. Think of it as a universal language (based on JSON RPC) that allows agents built on different frameworks (like Google’s ADK, CrewAI, etc.) to collaborate, much like how microservices talk to each other. It was created by Google, Microsoft, and Amazon to solve the problem of AI systems being “locked in” to a single framework.

2. Why is a “multi-agent system” better than a single, monolithic AI agent? A single, monolithic agent (like one big MCP) becomes inefficient, hard to manage, and difficult to scale as you add more functions. A multi-agent system is modular, like a team of specialists. Each agent focuses on one domain (e.g., one manages calendars, another manages bookings). This “microservice-style” approach makes the entire system more flexible, scalable, and easier to update.

3. What was the key achievement of this 3-4 week project? The key achievement was proving that interoperability between different AI frameworks is possible. In just 3-4 weeks, our developer built a system where agents running on two completely different platforms (Google ADK and CrewAI) successfully collaborated to complete a complex task. This demonstrates our ability to rapidly master and implement brand-new, cutting-edge protocols to build flexible, non-siloed AI solutions.

4. Can RUNNINGHILL build a custom multi-agent system for my business? Yes. This project is a direct proof-of-concept of that capability. We can analyze your business processes and design a “team” of specialized AI agents to automate them. Whether you need to integrate with calendars, databases, or third-party APIs, we can build a scalable, secure, and modular multi-agent system that fits your specific needs.

5. How does this project relate to your “Agents as a Service (AaaS)” capability? This project is a perfect example of how we build an AaaS platform for you. Our “Agents as a Service” offering isn’t a pre-built product; it’s our capability to design, build, and deploy a “digital workforce” that is custom-made for your company. We architect autonomous systems of multiple agents (like the ones in this demo) that work within your specific processes and integrate with your tools, freeing up your team to focus on strategy.