Gerhard G. StocKinger · Published on May 2, 2025

Large language models have demonstrated just how powerful language can be as an interface. The next step is to connect these models to corporate data, applications, and other agents in a controlled manner.

Two protocols, two tasks

The Model Context Protocol (MCP) standardizes how AI applications access tools and context. Agent-to-Agent (A2A) addresses collaboration among autonomous agents. Simply put: MCP connects an agent to capabilities; A2A connects agents to one another.

This enables scenarios in which an agent understands a request, retrieves data from a system, tasks a specialized agent, and feeds the result back into a business process.

Architecture Is Becoming a Management Issue

Open standards reduce the effort required for integration, but they do not automatically resolve issues related to permissions, data quality, and accountability. Companies should clarify the following early on:

• Which systems and actions can agents access?
• What identity and what rights do they use?
• How are actions logged and approved?
• When does a person have to confirm something?

Conclusion: Competition isn't decided solely by which model is the best. What matters most is how securely and flexibly companies orchestrate their data, tools, and agents.

This post is based on my LinkedIn post from May 2, 2025, and has been expanded for the blog.