AI Agents
An AI agent is an application that uses an AI model together with instructions, context, tools, and control logic to pursue a task. The model may decide what information to retrieve or which authorized tool to call, while the surrounding application enforces permissions, validation, and execution boundaries.
Agent loop
Goal → model decision → tool/retrieval → observation → next decision → final response/action
Not every LLM application needs an agent. A deterministic workflow is often preferable when the steps are known in advance.
Production controls
- Restrict tools and permissions.
- Validate tool arguments and returned data.
- Bound loops, retries, and execution time.
- Require approval for high-impact actions when appropriate.
- Trace tool calls and failures.
- Evaluate complete tasks rather than only model responses.
AWS note
AWS documentation now distinguishes Amazon Bedrock Agents Classic from Amazon Bedrock AgentCore. Check the target platform before following older agent tutorials.
Connected knowledge
LLMs → tool calling → AI security → Bedrock Agents and AgentCore.
