AWS re:Invent 2025 - Powering your Agentic AI experience with AWS Streaming and Messaging (ANT310)

Powering Agentic AI Experiences with AWS Streaming and Messaging

Overview

• This presentation explored how AWS streaming and messaging services can power agentic AI experiences, using a real-world use case from the Olympics.
• The speakers discussed the fundamental concepts of streaming, messaging, and agentic AI, as well as key design patterns and a live demonstration.

The Importance of Data and Streaming

• Organizations are moving forward with AI initiatives, but many struggle with data foundation challenges like team collaboration, data fragmentation, governance, and scalable infrastructure.
• Streaming data technologies can help address these challenges by:
  • Enabling real-time data capture and analysis
  • Reducing infrastructure costs by filtering out unnecessary data
  • Powering AI with fresh, high-quality data
  • Keeping systems in sync to support complex workflows
• Real-time data is the foundation for modern, agentic AI, which combines real-time decision-making, machine learning, and generative AI capabilities.

Streaming and Messaging for Agentic AI

• Streaming provides the real-time intelligence and closed-loop process needed for agentic AI, allowing continuous data gathering, processing, insight delivery, and autonomous action.
• Messaging services like Amazon SQS enable asynchronous communication between AI agents, providing scalability and elasticity to handle demanding AI workloads.

Agentic AI Use Cases

• Agentic AI is being used to power customer experiences, detect anomalies and provide predictive analytics in manufacturing, and reduce human bias in insurance claims processing.

Design Patterns for Real-Time Data Architectures

• Key components include:
  • Streaming storage (Amazon Kinesis, Amazon MSK) for durable, scalable data ingestion
  • Stream processing (Amazon Managed Streaming for Apache Flink) for real-time anomaly detection and enrichment
  • Agent deployment and management (Amazon Bedrock Agent Core) for building and scaling custom AI agents
• Messaging services like Amazon SQS enable asynchronous communication and coordination between specialized AI agents.

Olympics Use Case

• The Olympics faces massive scale and complexity, with 24,000 media personnel, 39 venues, 32 sports, and 91 countries winning medals.
• Digitally, the Olympics sees 325 million users and 16 billion social media engagements during the events.
• To handle this scale, the Olympics team uses a streaming-based architecture with:
  • 3.2 PB of user action data in Amazon Kinesis
  • 4.6 million play-by-play events from a legacy application, ingested via S3 and Lambda
  • Anomaly detection to identify "interesting moments" that can be summarized and delivered to content creators using agentic AI
• For the upcoming Milan Cortina Olympics, the team is evolving their architecture to use Amazon Bedrock Agent Core for more scalable, maintainable, and observable agentic AI applications.

Key Takeaways

1. Streaming data is a must-have for powering agentic AI experiences.
2. Data is the foundation that enables AI, and organizations need to address data challenges.
3. Messaging services like Amazon SQS enable asynchronous communication and coordination between AI agents.
4. Amazon Bedrock Agent Core provides a scalable, observable platform for building and deploying custom AI agents.
5. The Olympics use case demonstrates the real-world application of these technologies to handle massive scale and complexity.