Enterprise AI is moving beyond simple chatbots and static data. The next big thing is agentic AI — autonomous processes that can perceive, reason, and act across company systems. Instead of waiting for a prompt, these agents operate dynamically, coordinating with each other to get things done.

What Are Model Context Protocols and Agent2Agent Communication?

New standards like MCP (Model Context Protocol) and A2A (Agent2Agent) are generating excitement. MCP aims to be the universal connector for enterprise data, much like a USB-C port for data access. It’s open and already being adopted by hundreds of companies. A2A, on the other hand, is more ambitious. It focuses on making AI agents coordinate independently, rather than just connecting to systems.

However, both protocols have a big limitation: they are stateless. When an agent sends or receives a message, it forgets everything afterward. There’s no record of the interaction. This works fine for experiments but is a problem for real-world applications. Complex workflows, like financial planning involving multiple agents, need to remember past interactions to function smoothly and troubleshoot issues.

The Data Infrastructure Challenge for Agentic AI

Building effective agentic AI isn’t just about protocols. It’s a data architecture problem. Agents are decision-makers that react to their environment in real time. They need to access current, trustworthy data quickly. For example, an e-commerce platform deploying agents for inventory, customer service, and fraud detection must share context. If an inventory agent notices unusual demand, the fraud detection agent should know immediately. If a customer service agent resolves a complaint, other agents need to know that too.

Without a way to share this context, each interaction starts from scratch, making coordination difficult. That’s why event-driven design has become essential. Instead of simply calling APIs, systems now use event streaming technologies like Apache Kafka. Kafka creates a history of interactions, which agents can review, react to, and learn from. It acts as a shared memory, giving agents a reliable record of past events, decisions, and observations.

This setup turns stateless protocols into powerful, stateful systems. Kafka logs enable troubleshooting, auditing, and replaying interactions. They provide the durability and visibility needed for complex, enterprise-scale AI operations. In essence, Kafka’s event streams become the “memory layer” that protocols like MCP and A2A rely on for effective coordination.

Lessons from Microservices for Agentic Architectures

Looking back at the rise of microservices can teach us a lot. Early microservice efforts often failed because teams jumped into building platforms without understanding their infrastructure needs. They started with big, complex systems, expecting instant results. That approach led to delays and skepticism. Instead, successful microservice architectures began with foundational tools like Kafka, which allowed services to share state and communicate reliably.

The same principle applies to agentic AI. Start by establishing a robust data infrastructure before deploying protocols. Build a trustworthy data store where agents can discover and share information. Use event streaming to enable agents to communicate asynchronously, maintain context, and coordinate smoothly. Gradually add more complex interactions—like agent-to-agent coordination—once your infrastructure can support it.

Another key lesson is that focusing on data quality matters more than choosing the latest AI models. Even the most advanced language models can produce poor results if they operate on stale or incomplete data. Ensuring that your data is real-time, accurate, and accessible is critical for agent decision-making. When deploying MCP or A2A, prioritize building a strong data foundation that allows your agents to reason, react, and coordinate effectively.

In summary, the future of enterprise AI depends on blending protocols with solid data architecture. By learning from past tech transformations, organizations can better prepare for an agentic future—one where autonomous agents work seamlessly across systems, powered by reliable, real-time data streams.

Inspired by

• https://www.infoworld.com/article/4046484/beyond-ai-protocols-preparing-for-mcp-and-a2a-in-production.html

Sources

• gmpg.org
• modelcontextprotocol.io
• github.com
• podigee.com
• podigee.com