AI's Impact on Software Architecture Evolution: Explicit vs. Implicit Control Flow

The rise of AI coding assistants presents a fascinating challenge to software architecture evolution. AI models are highly proficient in established patterns, particularly those like Spring's annotation-driven dependency injection and implicit control flow, due to their vast presence in training data. This raises the question of whether AI's inherent bias towards familiar patterns could hinder the adoption of novel architectural approaches.

The Challenge of Implicit Architectures for AI

Traditional `@RestController` patterns in frameworks like Spring rely heavily on implicit control flow, where the order of operations (validation, business logic, auditing) is embedded within the call stack and method invocations. To understand or modify this flow, human developers must read and interpret code, as the wiring is not explicitly defined as data. This reliance on implicit behavior makes it challenging for AI, which excels at pattern recognition but struggles with inferring complex, unstated relationships across a codebase.

Function Injection: A Shift to Explicit Control Flow

The article proposes Function Injection as an architectural evolution, moving "what happens next" from the implicit call stack to explicit, configured data. Similar to how Dependency Injection externalized dependencies from imperative constructor code, Function Injection externalizes control flow. In the example provided, REST endpoints are defined in YAML files, where the sequence of functions (validate, build, audit) and conditional branches are declared, not buried in code. This creates an explicit

validate:
  class: ValidateGreetingLogic
  outputs:
    valid: build
build:
  class: PostGreetingLogic
  next: audit
audit:
  class: AuditGreetingLogic

The Role of Documentation in AI-Driven Architecture Adoption

The core finding is that AI is not inherently conservative. While it defaults to familiar patterns due to training data prevalence, it can readily adopt new architectures if they are clearly and completely specified. The bottleneck for adopting new architectural patterns with AI is often the lack of comprehensive and unambiguous documentation. If new approaches are well-documented with clear schemas, conventions, and examples, AI can quickly learn and apply them, shifting the burden of explicit configuration from human developers (who preferred code-adjacent annotations) to comprehensive documentation for AI.