AWS Builder Center
Build a Real AI Agent in 15 Minutes with Amazon Bedrock AgentCore Harness

Build a Real AI Agent in 15 Minutes with Amazon Bedrock AgentCore Harness

No orchestration code. No containers. Just configuration. Here’s how to go from zero to a working, production-ready agent using the managed AgentCore harness — the fastest path available right now.

Why this matters today

In the last few months Amazon Bedrock AgentCore has matured significantly. The managed harness is now generally available, Web Search is built-in, long-running sessions are supported, and continuous optimization features are rolling out.
More importantly, real companies are already using it in production (recent examples include AvioBook for airline operations and Heurist Finance for investment analysis). The barrier to entry has never been lower.
If you still think building an AI agent requires writing complex loops or managing infrastructure, this tutorial will change your mind.

What you will build

A simple but useful Research Assistant agent that can:
  • Answer questions using current web knowledge
  • Remember conversation context
  • Run securely on AgentCore Runtime
Time required: 10–15 minutes

Prerequisites

  • AWS account
  • AWS CLI configured
  • Node.js 18+ installed

Step 1: Install the AgentCore CLI

Bash
npm install -g @aws/agentcore
Verify:
Bash
agentcore --version

Step 2: Create a new project with the managed harness

Bash
agentcore create
When the wizard appears:
  • Choose Harness (this is the key difference from code-based agents)
  • Select a name (e.g. research-assistant)
  • Choose Amazon Bedrock as the model provider
  • Pick Claude Sonnet or Amazon Nova (both work great)
The CLI will generate a clean project with a harness.json (or equivalent config) file instead of Python/TypeScript agent code.

Step 3: Configure your agent (the fun part)

Open the generated configuration file. You mainly need to set three things:
  1. System prompt / Instructions Example: text __PRE_PLACEHOLDER_3__
  1. Model Use any model available in your region (Claude Sonnet 4 or Nova Pro are excellent defaults).
  1. Tools Enable the built-in Web Search tool (now generally available). This alone makes the agent significantly more useful.
You can also add memory with a single CLI command later:
Bash
agentcore add memory

Step 4: Test locally

Bash
agentcore dev
Then invoke it:
Bash
agentcore invoke '{"prompt": "What are the latest improvements in Amazon Bedrock AgentCore as of September 2026?"}'
You should get a grounded answer using current information.

Step 5: Deploy to production

Bash
agentcore deploy
That’s it. Your agent is now running on AgentCore Runtime with:
  • Session isolation
  • Automatic scaling
  • Built-in observability
  • Secure execution environment

Why the Harness is so powerful right now

Approach
Speed
Control
Best for
Managed Harness
Very fast
Medium
Most use cases
Code-based (Strands)
Medium
Full
Complex custom logic
Classic Bedrock Agents
Fast
Low
Simple legacy use cases
The harness lets you start simple and export to full code later if you need more control — without rewriting everything.

Next steps you can do this weekend

  1. Add long-term memory so the agent remembers user preferences.
  1. Connect your own APIs using AgentCore Gateway + MCP.
  1. Enable the new continuous optimization features (failure & trajectory insights).
  1. Turn this into a multi-agent system.

Final thoughts

In September 2026, the gap between “I want an AI agent” and “I have a working agent in production” has shrunk dramatically.
You no longer need to be an expert in agent frameworks to ship something useful. The AgentCore harness + Web Search combination is currently one of the fastest and most practical ways to get started.
Go build something. Then come back and share what you created in the comments — I’d love to see it.
Happy building!
Here’s a clean, ready-to-paste Study Track section that matches the style of the last part of your article. Just copy and drop it at the end.



Recommended Study Track

If you want to go deeper after finishing this tutorial, follow this progressive path:
1. Official Quick Starts (Do these first)
2. Core Documentation
3. Hands-on Workshops & Labs
4. Structured Learning (AWS Skill Builder)
5. Code & Samples
Any opinions in this article are those of the individual author and may not reflect the opinions of AWS.
Enjoyed reading this content? Let the author know!

Your likes, comments, shares, and saves help creators reach more builders.

Loading recommendations

Loading article