Welcome to the Deploying AI Agents on AWS workshop.
In this hands-on workshop, you’ll learn how to build, deploy, and connect AI agents on AWS using Pulumi and Amazon Bedrock AgentCore. By the end, you’ll have a multi-tool agent that browses the web, runs Python code, remembers user preferences, and writes reports to S3.
What you’ll build
flowchart TD
O["Orchestrator Agent<br/>Coordinates delegation"]
style O fill:#6366f1,color:#fff,stroke:#4f46e5,stroke-width:2px
O --> A["Basic Agent<br/>Q&A"]
O --> B["MCP Server<br/>JWT + Cedar Policies"]
O --> C["Specialist Agent<br/>Deep Analysis"]
O --> D["Weather Agent<br/>Full Stack"]
D --> E["Browser Tool"]
D --> F["Code Interpreter"]
D --> G["Memory"]
D --> H["S3 Results"]
style A fill:#22c55e,color:#fff,stroke:#16a34a,stroke-width:2px
style B fill:#22c55e,color:#fff,stroke:#16a34a,stroke-width:2px
style C fill:#22c55e,color:#fff,stroke:#16a34a,stroke-width:2px
style D fill:#22c55e,color:#fff,stroke:#16a34a,stroke-width:2px
style E fill:#f59e0b,color:#fff,stroke:#d97706,stroke-width:2px
style F fill:#f59e0b,color:#fff,stroke:#d97706,stroke-width:2px
style G fill:#f59e0b,color:#fff,stroke:#d97706,stroke-width:2px
style H fill:#f59e0b,color:#fff,stroke:#d97706,stroke-width:2px
Your agents will deploy as containerized services on AgentCore Runtime, authenticate with JWT tokens via Cognito, enforce access control with Cedar policies, and communicate agent-to-agent using IAM-scoped invocations.
Workshop chapters
| Chapter | Title | Duration |
|---|---|---|
| 00 | Setup and orientation | 15 min |
| 01 | Your first agent on AgentCore | 30 min |
| 02 | Hosting an MCP server behind an AgentCore Gateway | 45 min |
| 03 | Multi-agent orchestration | 40 min |
| 04 | The full stack: weather agent with tools and memory | 40 min |
| 05 | Cleanup | 10 min |
| Total | 180 min |
Speakers
| Name | Role | Company |
|---|---|---|
| Engin Diri | Senior Solutions Architect | Pulumi |
| Adam Gordon Bell | Community Engineer | Pulumi |
Prerequisites
- Laptop with internet access
- AWS account with Bedrock model access enabled (provided for the workshop)
- Pulumi account (free tier works)
- GitHub account
- TypeScript path: Node.js 18+ installed
- Python path: Python 3.11+ and uv installed
- Python 3.11+ for agent code and testing (both paths):
pip install boto3 mcp - Basic terminal familiarity
We recommend GitHub Codespaces for a zero-install experience. Click the badge at the top of this page.
| Machine type | Cores | RAM | Recommended |
|---|---|---|---|
| Standard | 4-core | 16 GB | Yes |
Troubleshooting
pulumi up hangs during CodeBuild: The first build takes 5-10 minutes while Docker images are built and pushed to ECR. This is normal.
AWS credentials expired: Run pulumi env open aws-bedrock-workshop/dev to verify your credentials are configured correctly, then retry.
Agent invocation returns 500: Check CloudWatch Logs at /aws/bedrock-agentcore/runtimes/ for your runtime. Common causes are missing IAM permissions or environment variables.
CodeBuild fails: Check the build logs in the AWS Console under CodeBuild > Build projects. The most common issue is ECR permission errors during docker push.
Weather agent test hangs: The first invocation triggers a cold start (1-2 min). The test script handles this. If it times out, run it again.