Working of the Bedrock Agent

Question

I understand that a Bedrock Agent has autonomy to decide which tools to use, how many times to use them, and what criteria determine task completion. From my understanding, this autonomy is coordinated by an agent prompt (sometimes referred to as the agent flow prompt), which is different from the orchestration prompts that Bedrock injects automatically.

My question is:

Is there a way to see the agent flow/coordination prompt that drives this decision-making process?

I am not referring to the orchestration prompts that manage tool calls or enforce structure, but specifically the prompt that governs the agent’s overall reasoning and task completion behavior.

This would help clarify how much of the agent’s behavior is controlled by the prompt I provide versus the hidden system-level coordination logic.

re:Post Agent · Accepted Answer

Amazon Bedrock Agents use an orchestration strategy to determine how they accomplish tasks, which involves developing a plan and executing it to reach a final answer. By default, agents use a strategy called ReAct (Reason and Action) defined in the base default prompt templates.

You have two main options to customize and potentially view the orchestration logic:

1. Advanced Prompt Templates: You can modify the base prompt templates that control the agent's behavior during pre-processing, orchestration, knowledge base response generation, and post-processing. This allows you to override the default logic with your own configurations.

2. Custom Orchestration: For more control, you can create an AWS Lambda function that outlines your own orchestration logic. This gives you full control over how your agent handles multi-step tasks, makes decisions, and executes workflows. The Lambda function controls when and how to invoke the model, when to invoke action tools, and how to determine the final response.

Additionally, you can use traces to examine your agent's reasoning process at each step of its orchestration, which can help you understand the decision-making process.

While you can customize these components, the underlying system-level coordination logic that drives the agent's autonomy is not directly exposed as a single viewable "agent flow prompt." Instead, you interact with it through the customization options mentioned above, which allow you to influence how the agent makes decisions about tool usage and task completion.  
**Sources**  
[How Amazon Bedrock Agents works - Amazon Bedrock](https://docs.aws.amazon.com/bedrock/latest/userguide/agents-how.html)  
[Customize your Amazon Bedrock Agent's behavior with custom orchestration - Amazon Bedrock](https://docs.aws.amazon.com/bedrock/latest/userguide/agents-custom-orchestration.html)  
[Automate tasks in your application using AI agents - Amazon Bedrock](https://docs.aws.amazon.com/bedrock/latest/userguide/agents.html)  
[Customize agent orchestration strategy - Amazon Bedrock](https://docs.aws.amazon.com/bedrock/latest/userguide/orch-strategy.html)

Working of the Bedrock Agent

Add your answer

Relevant content