ABOUT
What is reopt architecture
reopt architecture is a methodology for building AI products that do not collapse as they grow. It maintains ownership of outcomes, contracts for judgment, and evolution of structure as a system.
In an era when AI guarantees speed, scaling without structure produces agentic debt. It tells you why and how to build products from structure rather than from prompts.
Author
Contributing
This project will evolve into an open reference. Pattern proposals, case shares, terminology debates — every kind of contribution is welcome.
Methodological grounding
The assessment system in reopt architecture is built on validated methods from software engineering and AI governance. The eight engineering-perspective attributes are aligned with the ISO/IEC 25010 quality model. Trade-off analysis between attributes follows CMU SEI's ATAM (Architecture Tradeoff Analysis Method). The governance framework references Singapore IMDA's Model AI Governance Framework for Agentic AI and OpenAI's Practices for Governing Agentic AI Systems, adopting an AI Governance by Design approach that bakes governance into the design stage.
Relationship to existing methodologies
reopt architecture inherits insights from existing software architectures, extending them to the non-deterministic execution and governance requirements of AI agents.
DDD — Bounded Context
Shared: Bounded Context and Responsibility Boundary address the same problem: explicitly separating responsibility boundaries and structuring communication across them.
Difference: The DDD Bounded Context is a static boundary for domain-model consistency. The responsibility boundary in reopt architecture is designed for non-deterministic agent execution and is adjusted dynamically from operational data. DDD also does not treat governance as part of the architecture, while reopt architecture includes the governance layer as a baseline structural element.
마이크로서비스 — Service Contract
Shared: Service Contract and Module Contract share the same principle of defining contract-based interfaces between services.
Difference: Microservice contracts focus primarily on the input/output schema of the success path. The Module Contract in reopt architecture treats failure conditions, refusal logic, and authority scope as required elements. Because AI agents are not deterministic, "when to refuse" matters as much as "what to return."
Conway's Law
Shared: The observation that system structure mirrors organizational structure applies to agent systems too. The collaboration structure between agents becomes the system architecture.
Difference: Conway's Law describes the passive reflection of organizational structure into the system. reopt architecture flips this: by deliberately designing the agent organization, it determines the system structure. The Responsibility Partitioning pattern is the concrete method of that deliberate design.
Anthropic — Brain/Hands/Memory 분리
Shared: Both share the principle of separating the system into components that can evolve independently. Anthropic's three-way split — Session (memory), Harness (brain), Sandbox (hands) — corresponds structurally to reopt architecture's four-layer model.
Difference: Anthropic's separation comes from an infrastructure-operations perspective (container fault recovery, security isolation, deployment independence). The four layers in reopt architecture come from a governance perspective (attribution of responsibility, contract validation, collaboration coordination, policy enforcement). reopt architecture also breaks out Collaboration as its own layer, treating inter-agent handoff and context-passing rules as a separate design subject.
References
The patterns, model, and principles of reopt architecture reference and build on the insights in the following sources.
Anthropic Engineering
Agent-Computer Interface (ACI), Poka-yoke design, Evaluator-Optimizer pattern. Empirical grounding for the Module Contract and Evaluation patterns.
Anthropic Engineering
Scaling Managed Agents: Decoupling the Brain from the Hands
Session (memory) / Harness (brain) / Sandbox (hands) separation architecture. An empirical counterpart of the four-layer model.
Anthropic Engineering
Effective Harnesses for Long-Running Agents
Harness as governance infrastructure. The pattern of using a feature list as a contract. A concrete implementation of the Evolution stages.
Anthropic Engineering
Demystifying Evals for AI Agents
pass@k vs pass^k metrics, graduating capability eval to regression eval, eval-driven development. Core grounding for the Evaluation pattern and the SHARPEN phase.
Anthropic Engineering
Effective Context Engineering for AI Agents
Context rot, attention budget, progressive disclosure, compaction. Empirical grounding for Context Routing and State/Memory patterns.
Anthropic Engineering
How We Built Our Multi-Agent Research System
Orchestrator-Worker pattern, sub-agent delegation, source-quality heuristics. An empirical case for Responsibility Partitioning and the Collaboration layer.
Anthropic Engineering
Writing Effective Tools for Agents — with Agents
Tools as contracts with non-deterministic systems. Consolidated vs distributed design, prompt engineering of error responses. An implementation guide for the Module Contract.
Anthropic Engineering
Claude Code Auto Mode: A Safer Way to Skip Permissions
Three-tier authority model and four threat types (overeager, honest mistakes, prompt injection, misalignment). An implementation model for the Human Approval pattern.
Anthropic Engineering
Harness Design for Long-Running Application Development
Planner-Generator-Evaluator three-agent design, Sprint Contract (dynamic contract negotiation), criteria-based grading. An implementation case for the Module Contract and Evaluation patterns.
InfoQ
Google's Eight Essential Multi-Agent Design Patterns
Eight implementation patterns including Sequential Pipeline, Parallel Fan-out, and Hierarchical Decomposition. The basis for the Implementation Bridge mapping.
InfoQ
Google Publishes Scaling Principles for Agentic Architectures
Quantitative evidence that multi-agent composition does not always raise performance. Empirical grounding for Layer, Then Scale.
InfoQ — QCon AI NY 2025
Becoming AI-Native Without Losing Our Minds To Architectural Amnesia
Origin of the agentic-debt concept — debt that accrues when autonomy outpaces discipline. The basis for the Agentic Debt framing.
InfoQ
Agentic AI Architecture Framework for Enterprises
Foundation → Workflow → Autonomous three-stage framework. A reference for comparison with the Evolution stages.
InfoQ
The Architectural Shift: AI Agents Become Execution Engines While Backends Retreat to Governance
The argument that backends are retreating into a governance layer. An industry response to the Agents Scale by Structure thesis.
InfoQ
From Prompts to Production: a Playbook for Agentic Development
ASDLC (Agentic SDLC) concept. The principle of defining "what must never happen." The basis for positioning OCLS within ASDLC.
Carnegie Mellon SEI
ATAM: Method for Architecture Evaluation
Methodology for stratifying quality attributes via the Utility Tree and analyzing trade-offs. The academic basis for the assessment slider → tension-detection structure.
ISO/IEC
ISO/IEC 25010 — Systems and Software Quality Models
International standard of eight quality attributes (security, reliability, performance efficiency, etc.). The alignment basis for the eight-axis development perspective.
IMDA (Singapore)
Model AI Governance Framework for Agentic AI
Agentic-AI governance model framework. Responsibility assignment, contract-based control, layered oversight, operational adjustment — structurally aligned with the OCLS loop.
Preprints.org
AI Governance by Design for Agentic Systems
AIGD (AI Governance by Design) approach that bakes governance into the design stage. Proposes Constitutional AI, Compliance, Access Control, and Audit Trail pattern clusters.
World Economic Forum
AI Agents in Action: Foundations for Evaluation and Governance
Foundational framework for evaluating and governing AI agents. International consensus on multi-dimensional evaluation metrics and governance principles.
OpenAI
Practices for Governing Agentic AI Systems
Practitioner guidelines for safe operation of agentic AI. Technical controls and processes across the AI lifecycle.
arXiv
Architecting Agentic Communities using Design Patterns
Required application of Validation, Human-in-the-Loop, and Audit Trail patterns in regulated industries (HIPAA, SEC/FINRA). Academic basis for the financial-services preset profile.
arXiv
Beyond Task Completion: An Assessment Framework for Evaluating Agentic AI Systems
Multi-dimensional evaluation of reasoning, planning, collaboration, and ethical alignment beyond task completion. Aligned with the multi-dimensional approach of the integrated assessment metrics.
Future Internet (MDPI)
The Rise of Agentic AI: Architectures, Taxonomies, and Evaluation Metrics
Comprehensive review of agentic-AI architectures, taxonomies, and qualitative/quantitative evaluation metrics. A reference for evaluation-framework design.