The central concept is to treat architectural definitions and constraints as executable code rather than static documents or diagrams. This is primarily achieved through 'architecture fitness functions'—automated tests that verify architectural characteristics, ensuring the implementation aligns with the intended design.
The speakers are developing a new, platform-agnostic ADL to describe architectural rules and structures. This language is designed to be a precise, human-readable pseudocode that captures architectural intent without being tied to a specific implementation technology.
A key innovation is using the ADL as a high-level input for Large Language Models (LLMs). The ADL is designed to be 'prompt friendly,' enabling LLMs to act as a translation layer, generating concrete, executable fitness functions in various programming languages and testing frameworks.
The speakers explicitly differentiate their approach from past attempts like MDA. Unlike MDA, which tried to generate entire applications from models, 'Architecture as Code' focuses on verifying high-value constraints (the 'load-bearing walls') rather than generating application code, making it a more agile and targeted approach.
Keep pulling the thread on Neal Ford & Mark Richards.