How to Build Claude AI Agents with Amazon Bedrock

TL;DR

Build a Claude-powered agent by installing Strands Agents, selecting a model, supplying tools, and writing a prompt that defines the objective. Amazon Bedrock provides managed model access through a unified API, while Strands lets the model reason, plan, invoke tools, evaluate results, and continue until it completes the requested task.

Transcript

Building AI agents with Claude in Amazon  Bedrock. Today I am excited to explore how to create intelligent autonomous AI systems that  can transform your applications. My name is Dwan Lightoot. I am a developer advocate at AWS.  I'm Banjo. I'm a systems architect at AWS. Hi everyone. My name is Suman DNA. I'm a developer  advocate at AWS. Now, this... Read More

Key Insights

• Amazon Bedrock is a fully managed service that provides access to foundation models, including models in the Claude family, through a unified API. It also supplies model choice, guardrails, enterprise-grade security by default, and infrastructure for scaling AI applications globally.
• An AI agent is an autonomous system that can reason, plan, and take multiple steps to accomplish an objective. It converts a high-level task into a plan, acts on the plan, evaluates intermediate results, and decides what to do next until the objective is achieved.
• Strands Agents is an open-source SDK designed around three primary components: a model, tools, and a prompt. Its simplified structure gives the underlying language model flexibility to reason without requiring extensive scaffolding for backstories, goals, or separately structured prompt elements.
• Claude 3.7 was the default model used by Strands at the time of the workshop. The SDK can also work with different model options, including models accessed through Amazon Bedrock, LiteLLM, and Llama, while the demonstrated code switched to Claude 3.5 for speed.
• Strands includes built-in tools that reduce the amount of supporting code developers need to write. In the weather example, an existing HTTP request tool calls a government weather endpoint, while a separate word-count tool processes the generated response.
• Custom Strands tools can be created by defining a function and applying the tool decorator. The demonstrated approach allows developers to concentrate on function behavior and return values while the framework handles the additional integration needed for agent tool use.
• Claude Code can operate through Amazon Bedrock without an Anthropic key in the demonstrated AWS environment. After exporting the specified Claude Code environment variable, the workshop environment was already configured to use Bedrock and could proceed with the recommended settings.
• Strands applications can move beyond local testing and be deployed to AWS compute services. The presenters specifically identify EC2, Lambda, and ECS as supported deployment environments, making it possible to place agent code alongside an existing cloud workload.

Questions & Answers

Q: What is an AI agent in Amazon Bedrock?

An AI agent is described as an autonomous system that can reason, plan, and take multiple steps to complete an objective. After receiving a high-level task, it determines the steps required, takes actions, evaluates the resulting information, and reasons about what should happen next. This cycle continues until the agent achieves the requested objective.

Q: How do you build a Claude agent with Strands Agents?

A Strands agent is built by defining three central elements: a model, a set of tools, and a prompt. The prompt contains the request or objective, while the agent object identifies the model and tools it can use. Once configured, the model can reason about the request, select tools, process their results, and continue toward completion.

Q: What does Amazon Bedrock provide for Claude applications?

Amazon Bedrock provides managed access to foundation models, including models from the Claude family, through a unified API. The service is presented as supplying model choice, guardrails, enterprise-grade security by default, and support for scaling AI applications globally. In the workshop, Bedrock also supplies the model access required by Strands Agents and Claude Code.

Q: How do you install Strands Agents for the workshop?

The workshop installs the Strands Agents framework and its tools package with pip, using the packages identified as Strands Agents and Strands Agents Tools. It also uses the UV utility for downloading MCP servers. After installation, developers can create an agent object, select a model, add available tools, and send a prompt to the agent.

Q: How does the Strands weather and word-count example work?

The example gives the agent a system prompt instructing it to find weather information and present it in a human-readable form. A built-in HTTP request tool queries the weather.gov endpoint without an API key. The agent also receives a word-count tool, allowing it to count the words in the response and demonstrate the coordinated use of multiple tools.

Q: How are custom tools created in Strands Agents?

A custom Strands tool is created by defining a function, specifying its behavior and return value, and adding the tool decorator. The presenter emphasizes that this avoids extensive integration code around the function. The tool can then be included when the agent object is created, making it available for the model to select during its reasoning process.

Q: Can Claude Code use Amazon Bedrock without an Anthropic key?

Claude Code can use Amazon Bedrock in the demonstrated AWS setup without requiring an Anthropic key or a separate Anthropic sign-in. The workshop enables this behavior by exporting the specified Claude Code environment variable. Because the provided environment already contains the necessary AWS and Bedrock configuration, Claude Code becomes ready after the initial settings are accepted.

Q: Where can a Strands agent be deployed on AWS?

A Strands agent can be deployed to AWS rather than remaining limited to a local test environment. The presenters specifically mention EC2, Lambda, and ECS as deployment targets with integrated support. This allows developers to run the same agent code in cloud environments where their workforce or application workloads are already hosted and connect it with Bedrock-hosted models.

Summary & Key Takeaways

• Amazon Bedrock is presented as a fully managed service for accessing foundation models, including the Claude family, through a unified API. It supports building and globally scaling AI applications with model choice, guardrails, and enterprise-grade security available by default, while also enabling Claude Code access through an AWS environment.

• AWS defines an agent as an autonomous system that can reason, plan, and take multiple steps toward an objective. The system interprets a high-level task, creates a plan, performs actions, evaluates the results, and determines what should happen next until it reaches the intended outcome.

• The workshop introduces the open-source Strands Agents SDK, whose core requirements are a model, tools, and a prompt. Participants install the agent and tool packages, enable supported Claude models in Amazon Bedrock, and examine an example that retrieves weather data and counts words using built-in and custom tools.