Architectural Patterns

Patterns are reusable architectural designs and structural solutions to common problems. If agentic architecture is the conveyor belt for knowledge work, patterns are the factory blueprints. They describe the topology and relationships between agents, systems, and human operators.

Start Here

Three foundational patterns that define the core architecture.

Live Patterns

• Agent Constitution Persistent, high-level directives that shape agent behavior and decision-making before action.
• Constitutional Review Verification pattern that validates implementation against both functional requirements (Spec) and architectural values (Constitution).
• Experience Modeling The practice of treating the Design System as a formal schema that agents must strictly follow, preventing UI hallucinations.
• Model Routing Strategic assignment of LLM models to SDLC phases based on reasoning capability versus execution speed.
• Product Vision A structured vision document that transmits product taste and point-of-view to agents, preventing convergence toward generic outputs.
• Ralph Loop Persistence pattern enabling autonomous agent iteration until external verification passes, treating failure as feedback rather than termination.
• The ADR A structural pattern for capturing architectural decisions with context, rationale, and consequences in an immutable record.
• The PBI A transient execution unit that defines the delta (change) while pointing to permanent context (The Spec), optimized for agent consumption.

Experimental & Draft

• Agent Optimization Loop The recursive process of using feedback from scenarios to continuously tune agent prompts, context, and tools.
• Agentic Double Diamond A computational framework transforming the classic design thinking model into an executable pipeline of context verification and assembly.
• Context Gates Architectural checkpoints that filter input context and validate output artifacts between phases of work to prevent cognitive overload and ensure system integrity.
• Context Map A high-density navigational index that enables agents to locate knowledge without managing massive context windows.
• External Attention External Attention offloads document processing to isolated sub-agents, returning only extracted answers to the main agent's context window.
• Spec Reversing Using frontier models to derive specifications from existing code to bootstrap the Agentic SDLC in brownfield projects.

Next Steps

• Practices — See patterns applied in real workflows.
• Getting Started — Deploy the methodology from scratch.
• Resources — Field manual, design system, and governance.