Randy DeFauw

Explore how chaos engineering (resilience testing) principles apply to autonomous AI agent workflows in this chalk talk. Learn to stress-test agent-based systems that handle complex, multi-step tasks using AWS Fault Injection Service.

In this session, you’ll see how to build and deploy a GenAI agent and Model Context Protocol (MCP) with Databricks, Anthropic, Mosaic External AI Gateway, and Amazon Bedrock. You will learn the architecture, best-practices of using Databricks Mosaic AI, Anthropic Sonnet 3.7 first-party frontier model, and LangGraph for custom workflow orchestration in Databricks Data Intelligence Platform. You’ll also see how to use Databricks Mosaic AI to provide agent evaluation and monitoring. In addition, you will also see how inline agent will use MCP to provide tools and other resources using Amazon Nova models with Amazon Bedrock inline agent for deep research. This approach gives you the flexibility of LangGraph, the powerful managed agents offered by Amazon Bedrock, and Databricks Mosaic AI’s operational support for evaluation and monitoring.

Resilience plays a pivotal role in the development of any workload, and generative AI workloads are no different. There are unique considerations when engineering generative AI workloads through a resilience lens.

There are multiple factors that impact the performance and reliability of generative AI applications. These include the choice of model, where the model is hosted, and the length and complexity of the inputs and outputs.

Docker containers and container technologies have become indispensable tools when scaling machine learning. However, many engineers and data scientists often struggle with understanding how these cloud-native technologies work, why they should use them, and how they can get started.