Machine Learning
Agents for Amazon Bedrock
Amazon Bedrock Agents Classic orchestrates foundation models, action groups, knowledge bases, session context, and legacy supervisor/collaborator teams. AWS placed the feature in maintenance mode for new customers on July 30, 2026; existing customers can continue using it while new designs should evaluate AgentCore.
Key points
- Action groups describe operations through functions or API schemas
- Agent aliases point applications to prepared versions
- Trace output helps inspect orchestration steps and tool decisions
- Legacy multi-agent collaboration uses a supervisor or supervisor-router with specialist collaborators
When to use it
- Build a support agent that retrieves policy and opens a case
- Coordinate read-only account lookup before drafting a response
Exam tips
- Treat Agents Classic scenarios as legacy architecture knowledge; choose AgentCore for a new flexible production agent platform unless a question states an existing deployment
- Validate every argument and require approval before irreversible or externally visible actions