Agent gateways make access control mandatory

On July 1, 2026, AWS described a serverless A2A gateway for agent discovery, routing and access control. This is not just an architecture pattern. Once multiple agents communicate with each other, companies need a reliable control layer between business applications, agents and tools.

What changes in practice

The AWS example places a central entry point in front of agents that may run on AWS Lambda, Amazon ECS, Amazon Bedrock AgentCore Runtime, other clouds or hybrid environments. According to AWS, Amazon Cognito handles the OAuth 2.0 client credentials flow, API Gateway and Lambda authorize requests, and DynamoDB stores the agent registry, permissions and rate limits.

The important point is the purpose of the gateway. It does more than route traffic. It uses JWT scopes to decide which calling application may reach which agent, caches Agent Cards and supports Server-Sent Events for streaming responses. Agent communication therefore becomes an operations and security topic, not a prompt-engineering topic.

Why A2A matters

The A2A project describes Agent2Agent as an open protocol for communication between separate agentic applications. Its core elements include JSON-RPC 2.0 over HTTP(S), discovery through Agent Cards, synchronous responses, streaming and asynchronous push notifications.

For CIOs, the consequence is more important than the specification itself. When agents publish capabilities, those capabilities must be inventoried, authorized, versioned and observed. Otherwise a new form of shadow IT emerges — this time with systems that can take action.

What DACH companies should check

For organizations in German-speaking markets, A2A gateways belong in the same discussion as API management, IAM and data protection. Teams should clarify early which agents may be reachable, which scopes make business sense, where personal data may appear and how tool calls are logged.

The pragmatic starting question is: do agents already have a central policy layer, or does security still depend on individual prompts and local integrations? Companies that want to use agentic AI in production should standardize discovery and access first. Scaling the agent landscape should come after that.