Bedrock Agents: Building Multi-Agent AI Systems on AWS

Building Autonomous AI Agents with Amazon Bedrock in 2025

Tech professional at work
Tech professional at work

AI agents are the next frontier. At re:Invent 2025, AWS announced major upgrades to Bedrock Agents, including multi-agent collaboration and the ability to run autonomously for days. Here’s how to build production-ready agents.

What’s New in Bedrock Agents

  • Multi-agent collaboration: Agents can delegate tasks to specialized sub-agents
  • Extended autonomy: Agents can work independently for hours or days
  • Nova Act: Browser automation with 90%+ reliability
  • DevOps Agent: Autonomous incident response

Creating a Bedrock Agent

import boto3

bedrock_agent = boto3.client('bedrock-agent')

# Create an agent
response = bedrock_agent.create_agent(
    agentName='customer-support-agent',
    foundationModel='amazon.nova-pro-v1:0',
    instruction="""
    You are a customer support agent. You can:
    1. Look up order status
    2. Process returns
    3. Answer product questions
    Always be helpful and professional.
    """,
    idleSessionTTLInSeconds=600
)

agent_id = response['agent']['agentId']

Adding Action Groups (Tools)

# Define Lambda function for order lookup
def lambda_handler(event, context):
    action = event['actionGroup']
    api_path = event['apiPath']
    parameters = event['parameters']

    if api_path == '/orders/{orderId}':
        order_id = parameters[0]['value']
        # Look up order in database
        order = get_order(order_id)
        return {
            'response': {
                'actionGroup': action,
                'apiPath': api_path,
                'responseBody': {
                    'TEXT': {'body': json.dumps(order)}
                }
            }
        }

# Attach action group to agent
bedrock_agent.create_agent_action_group(
    agentId=agent_id,
    actionGroupName='OrderManagement',
    actionGroupExecutor={
        'lambda': 'arn:aws:lambda:us-east-1:123456789:function:order-lookup'
    },
    apiSchema={
        's3': {
            's3BucketName': 'my-schemas',
            's3ObjectKey': 'order-api.json'
        }
    }
)

Multi-Agent Architecture

Supervisor Pattern

Create a supervisor agent that delegates to specialized agents:

  • Supervisor: Routes requests, manages workflow
  • Research Agent: Searches knowledge bases
  • Action Agent: Executes transactions
  • QA Agent: Validates responses

AWS DevOps Agent

One of the most exciting announcements: an autonomous on-call engineer that:

  • Monitors CloudWatch alarms
  • Analyzes logs across services
  • Correlates data from GitHub, ServiceNow
  • Identifies root causes
  • Suggests or implements fixes

This represents the future: AI agents that can manage infrastructure autonomously while humans focus on strategy.

Marcus Chen

Marcus Chen

Author & Expert

Marcus is a defense and aerospace journalist covering military aviation, fighter aircraft, and defense technology. Former defense industry analyst with expertise in tactical aviation systems and next-generation aircraft programs.

27 Articles
View All Posts