Building Agentic AI for Amazon RDS for SQL Server with Strands and AgentCore

The T-SQL diagnostic scripts that database administrators (DBAs) have accumulated over the years can become the foundation for autonomous AI agents through Amazon Bedrock AgentCore Runtime and Strands Agents. This allows existing knowledge to be transformed into agentic AI that understands context and performs complex tasks.

An agent reasons and plans through AI, and generates a final response through the agent loop (reasoning → tool selection → tool execution). A Tool is a function that allows an agent to interact with external systems, and in Strands, it is defined as a Python function with the @tool decorator applied. Wrapping the diagnostic scripts used by DBAs into Python functions turns them into tools for the agent.

This guide covers the process of building an agent that investigates Deadlock and Blocking issues occurring in Amazon RDS for SQL Server.

• Deadlock: Capture information via Trace flag 1204/1222 or the system_health Extended Event.
• Blocking: Use DMVs such as sys.dm_exec_requests to identify head blockers and waiting sessions.

DBAs can wrap existing diagnostic scripts into Python functions, define the agent's system prompt, and deploy it to AgentCore Runtime to build an autonomous database operations environment.