Amazon Bedrock Agents

Amazon Bedrock Agents is an AWS-managed service for building AI agents that can perform multi-step tasks by coordinating foundation models, enterprise data, APIs, software applications, and user conversations. Rather than requiring developers to manually program every step of an AI workflow, an agent can interpret a user’s request, break the task into smaller steps, retrieve relevant information, and call APIs or other tools to complete the requested action.

Amazon Bedrock Agents was first introduced in preview in July 2023 and became generally available in November 2023. At general availability, AWS positioned the service as a fully managed way to create generative-AI applications capable of executing multi-step tasks across company systems and data sources.

A core capability is agent orchestration. Developers provide natural-language instructions and select a foundation model, while the agent uses the model’s reasoning capabilities to create an execution plan. It can determine what information it needs, which action groups or APIs to call, and the order in which those operations should occur.

Action Groups

Action groups allow an agent to perform actions in external systems. Developers can define the parameters an agent needs to collect and the APIs it can call. AWS Lambda can be used to execute business logic, while Return of Control allows developers to take control of an action and implement their own application logic.

Knowledge Bases and RAG

Agents can be connected to Amazon Bedrock Knowledge Bases so they can retrieve information from organizational data when responding to users. This provides a Retrieval-Augmented Generation (RAG) approach in which relevant information is retrieved from a knowledge base and supplied to the foundation model to improve the response.

Foundation Model Support

Agents can use supported foundation models available through Amazon Bedrock. This allows developers to choose an appropriate model for their application rather than being locked into a single model provider. AWS states that Amazon Bedrock Agents supports the models supported by Amazon Bedrock, subject to regional availability.

Memory

Amazon Bedrock Agents supports memory retention so an agent can maintain relevant information across interactions. This can provide greater continuity for multi-step tasks and more personalized conversations.

Multi-Agent Collaboration

The service supports multi-agent collaboration, allowing multiple specialized agents to work together under a supervisor agent. A complex task can be divided into smaller tasks and assigned to specialist agents, which can then work together to produce a final result.

Code Interpreter

Agents can use a code interpreter to generate, execute, test, and troubleshoot code in a controlled environment. This enables use cases involving data analysis, calculations, and other tasks that benefit from programmatic processing.

Guardrails

Amazon Bedrock Agents can use Amazon Bedrock Guardrails to apply safety and policy controls to agent interactions. Guardrails can be associated with agents to help control model behavior and the information processed by the application.

Conversational Builder

The service provides a Conversational Builder in the Amazon Bedrock console. Developers can describe the purpose of an agent and the information it needs using natural language, and the builder can help configure the agent.

Inline Agents

AWS introduced InlineAgents in November 2024. This capability allows developers to configure foundation models, instructions, action groups, guardrails, and knowledge bases dynamically at runtime rather than creating a permanently configured agent first.

Deployment

After an agent is configured and tested, developers can create an alias for a version of the agent and invoke it from their applications through the Amazon Bedrock Agent Runtime API.

Current Product Status

Amazon Bedrock Agents is now designated Amazon Bedrock Agents Classic. AWS states that it is no longer open to new customers as of July 30, 2026. Existing customers can continue using the service, while AWS recommends Amazon Bedrock AgentCore for capabilities similar to Agents Classic.

This means businesses evaluating the AWS agent platform today should generally consider Amazon Bedrock AgentCore rather than treating Bedrock Agents Classic as a new standalone service.

Pricing

Amazon Bedrock Agents does not have a simple standalone per-user subscription price. Its costs are primarily associated with the underlying Amazon Bedrock model inference and other AWS services used by the agent.

Amazon Bedrock model pricing varies according to the model provider, model, modality, and service tier. AWS currently offers pricing tiers including Standard, Flex, Priority, and Reserved.

Agents may also generate costs from related services such as Knowledge Bases, model inference, data storage, API calls, AWS Lambda, and other AWS resources used by the application.

AWS also introduced Provisioned Throughput support for Bedrock Agents. Provisioned Throughput allows customers to purchase model units providing guaranteed input/output throughput and charges for those units by the hour, with different commitment options.

Pricing Model

Paid

Pricing Note

Amazon Bedrock Agents uses AWS consumption-based infrastructure and model pricing rather than a fixed per-user subscription. The total cost depends on the foundation models selected, token usage, agent activity, knowledge-base usage, and other AWS services used in the solution. Provisioned Throughput is also available for supported workloads.

Because Agents Classic is no longer open to new customers, current customers should also evaluate Amazon Bedrock AgentCore when considering AWS’s current agent platform.

Review

Amazon Bedrock Agents was a strong option for developers who wanted to build task-oriented AI agents while relying on AWS to manage much of the underlying orchestration infrastructure. Its ability to connect foundation models with APIs, knowledge bases, and enterprise systems made it suitable for applications that need to do more than generate text.

One of its biggest strengths is orchestration. Instead of developers manually implementing every step, the agent can determine which actions and information sources are needed to complete a request.

The combination of action groups and Knowledge Bases is particularly useful. Agents can both retrieve information and take actions, allowing developers to create applications that can answer questions and perform business operations within the same workflow.

Multi-agent collaboration also makes the platform suitable for more complex workflows in which different specialist agents can handle different parts of a task.

Another advantage is AWS’s broad AI ecosystem. Developers can choose from supported foundation models and combine agents with other Amazon Bedrock capabilities and AWS services.

The main disadvantage today is the product’s status. Agents Classic is no longer open to new customers, so it is not an appropriate choice for a new deployment. AWS now directs customers toward Amazon Bedrock AgentCore for similar agent-building and operating capabilities.

Pricing can also be difficult to predict because there is no single fixed subscription covering the complete solution. Model inference, knowledge bases, Lambda functions, storage, APIs, and other AWS services can each contribute to the final bill.

For organizations that already adopted the service, Amazon Bedrock Agents remains a capable framework for orchestrating AI-powered applications. For new AWS projects, however, Amazon Bedrock AgentCore is the more relevant current offering.