AWS re:Invent 2025 - Concept to campaign: Marketing agents on Amazon Bedrock AgentCore (AIM395)

Summary of AWS re:Invent 2025 Presentation: "Concept to campaign: Marketing agents on Amazon Bedrock AgentCore (AIM395)"

Challenges in Multi-Channel Marketing Automation

• Orchestrating complex customer journeys and workflows across multiple channels
• Delivering real-time personalized messaging and content at scale
• Unifying customer data from disparate systems and sources
• Achieving speed to market and campaign agility
• Accurately measuring attribution and ROI across touchpoints

Evolution of Agentic AI

• Progression from rule-based systems to fully autonomous agents
• Agentic AI agents: Autonomous software systems that can independently plan, act, and interact with the environment
• Benefits of agentic AI: Flexibility, real-time decision making, brand compliance, unified data view, speed to market, and advanced attribution

Amazon Bedrock AgentCore Services

1. Agent Core Runtime: Fully managed service to run AI agents and MCP servers
2. Agent Core Identity: Secure authentication and authorization for agents to interact with external services
3. Agent Core Gateway: Secure integration with existing APIs and Lambda functions, exposing them as MCP-compatible tools
4. Agent Core Memory: Provides short-term and long-term memory capabilities for agents
5. Agent Core Observability: Offers a single pane of glass for monitoring and auditing agent activities

Architectural Patterns for Agentic Systems

1. Agent-to-Agent Communication:

   • Options: AWS Boto3 APIs or A2A (Agent-to-Agent) protocol
   • Considerations: Native AWS features vs. open-source standards, integrating with legacy agents

2. Security:

   • Authentication and Authorization: IAM or OAuth
   • Propagating user identity to downstream MCP servers
   • Fine-grained access control within agent code

3. Multi-Tenancy:

   • Deployment options: Separate agent instances, versioned endpoints, or single endpoint with session isolation
   • Cost attribution and observability for multi-tenant environments

4. Observability:

   • Tracing agent behavior, decisions, and interactions
   • Integrating with existing observability tools using open standards

Epsilon's Agentic Platform Implementation

• Leveraging Bedrock AgentCore services to build a suite of marketing agents:
  • Campaign brief agent, branding agent, content creation agent, campaign execution agent, and performance analytics agents
• Benefits observed:
  • 30% reduction in campaign setup time
  • 20% increase in personalization capacity
  • Faster prototyping and deployment of new agents
• Lessons learned:
  • Shifting focus from workflows to outcome-based journeys
  • Rapid prototyping and experimentation with Bedrock AgentCore
  • Integrating agent-based pipelines with SDLC and operations

Future Directions

• Expanding agent-based automation across the organization (e.g., data, loyalty, operations)
• Building self-healing, self-monitoring agents
• Integrating a unified intelligence layer using Agent Core Gateway
• Commercializing and accelerating agent-based solutions for clients

Key Takeaways

• Agentic AI enables flexible, real-time, and scalable marketing automation
• Amazon Bedrock AgentCore provides a comprehensive set of services to build, deploy, and manage agentic systems
• Architectural patterns around agent communication, security, multi-tenancy, and observability are crucial for enterprise-grade agentic platforms
• Epsilon's experience demonstrates the business impact and lessons learned in adopting agentic AI for marketing automation