AWS Startups x JAM: Build, Deploy and Operate Agents with AWS

Join us for a hands-on half day where you'll take AI agents from idea to production on AWS. You'll have a working agent running in the first hour, then spend the rest of the morning adding what production agents need.

What you'll learn

Most agents stall between the demo and production. Building the agent loop is quick. Hosting it for many users, connecting it safely to your APIs, limiting what it can do, seeing what it did and knowing whether it's any good is where teams lose weeks.

In this workshop you'll use Amazon Bedrock AgentCore to do each of those things in minutes. You'll define an agent with a system prompt and a set of tools, and AgentCore will run it, isolate each user's session and scale it for you, with no servers to size or manage. Then you'll put your APIs behind a gateway, set hard limits the model can't argue its way past, trace every step, and score answers with evaluations. You'll also see how to use these capabilities with an agent you already run elsewhere, without moving it.

What to expect

• Introduction. What AgentCore is, how its pieces fit together, and how to tell which ones you need.
• Setup and access. A ready-made AWS environment with everything installed.
• Use case 1: Kopi Run. An office ordering agent that takes orders like "3 kopi-o siew dai and whatever Marcus had last time". You'll deploy it with the AgentCore harness, connect it to an existing API through AgentCore Gateway, track down a bug using its traces, and send it a rush of orders to see it scale with nothing to provision.
• Use case 2: Tally. A billing support agent that already runs on its own infrastructure and has been handing out credits it shouldn't. Without moving it, you'll route its tools through Gateway, cap what it can refund with AgentCore Policy, and use AgentCore Observability to compare what it did before and after.
• Use case 3: Intake. A finance agent that confidently repeats a misread invoice amount. You'll score its answers with AgentCore Evaluations, write an evaluator that knows your business, make the agent more careful with a configuration change, and set up evaluation that scores live traffic as it happens.
• One-to-one consults. Stay afterwards for time with AWS solutions architects and JAM engineers on your own use case.

You'll be working with provided code and configuration, so you can focus on understanding what each piece does. AI coding assistants such as Kiro are welcome but not required.

Who this is for

Technical founders, CTOs and lead engineers at startups who:

• have an agent idea or prototype that hasn't reached production yet, or
• already run an agent and need guardrails, observability or a way to measure quality.

Prerequisites

• Comfortable running commands in a terminal. Being able to read Python helps.
• A laptop with a current browser. A workshop AWS account and code editor are provided.
• No prior experience with Amazon Bedrock AgentCore needed.