Bayer adds AI agents to explain analytics workflows

The October 23 case study describes a supervisor directing questions about workflow failures or results to specialist agents. Their production tools have read-only access scoped to their responsibilities. Bayer saves workflow checkpoints in DynamoDB so work can resume across restarts or session boundaries. These are Bayer and AWS’s implementation claims, not independent test results.

The runtime documentation explains the storage distinction: by default, memory and files in the temporary execution environment survive only its compute lifecycle. Configured session storage can preserve a mounted directory across stops and resumptions; AgentCore Memory serves longer-lived structured information such as conversation history. Context can be reused across related calls with the same session identifier. However, the service does not enforce which user owns a session: the application backend should maintain that mapping and manage each user’s session lifecycle.

The gateway guide explains the other connection: an OpenAPI description turns an existing web API into an MCP-compatible tool, an interface an agent can call. The gateway translates between the two formats and combines MCP targets into one tool list, with semantic search for relevant tools. Its direct HTTP targets instead forward traffic without that aggregation or search. Client access to the gateway and the gateway’s credentials for downstream services are separate configuration concerns; connecting a tool does not erase those authorization boundaries.