AI-Friendly Architecture

AI impacts the structure of our programs in two ways. First, the program must be designed to work well in an AI-assisted software engineering environment. The blast radius of a change must be contained, for example, and the "context" must be managed to improve accuracy and minimize the cost of using the LLM to generate or change code. Next, the systems that we create that have AI embedded in them have to be secure, fault tolerant, and structured to ease agent-to-agent communication. In this session, you'll learn how to architect systems that support one or both of these goals.

You'll learn about:

• LLM Contexts and context management
• Micro-component and micro-Agent architectures that allow for incremental changes as we work with the LLM.
• Test-friendly architectures for LLM-generated applications.
• Building both REST and lightweight-messaging APIs for both human and agent-to-agent use.
• Agent-to-Agent communication strategies and the Model Context Protocol