Legacy services need agentic overlays

AWS has published a current pattern for connecting existing REST services to agentic architectures through agentic overlays. The important point for enterprises is not a full rebuild of stable business systems. It is an additional layer that makes REST endpoints usable for Agent-to-Agent communication and MCP-compatible tool access.

What the pattern proposes

AWS notes that many production enterprise architectures are built around REST APIs and microservices. These systems are reliable, but they were not designed for Agent-to-Agent communication. The proposed pattern places a thin wrapper layer in front of existing services: it receives agentic messages, translates them into existing API calls, and exposes functions as tools.

Google had introduced A2A as an open protocol for interoperability between agents. A2A is intended to let agents from different vendors and platforms collaborate in a coordinated way; MCP complements that world with tools and context. For enterprise architecture, this creates a new integration question: not every application has to become an agent, but it must be usable by agents in a controlled way.

Why this matters for CIOs

The approach is commercially relevant because it reframes modernization. Many core processes live in ERP, CRM, ticketing, or custom systems. A big-bang rewrite is expensive, risky, and difficult to justify organizationally. An overlay can instead show, step by step, which functions an agent may use, which data flows, and where authentication, logging, and cost controls must apply.

AWS also points to Amazon Bedrock AgentCore Gateway as a managed entry point. The documentation describes a gateway that converts APIs, Lambda functions, and services into MCP-compatible tools, fronts A2A traffic, and bundles authentication and observability. That is not a compliance guarantee, but it shows where platform providers are taking enterprise agents: away from isolated chat windows and toward controlled integration layers.

What DACH companies should examine

For DACH organizations, the leading question is not: “How quickly can we build agents?” It is: “Which existing services may be accessed agentically, under which permissions, and with which audit trail?”

A small start is sensible: one clearly scoped process, existing REST interfaces, technical policies, logging, and one owner for exceptions. Without these control points, agentic AI quickly becomes shadow integration. With them in place, it can modernize legacy systems without replacing them too early.