The AI-Native Continuous Loop

The AI-Native Continuous Loop A workflow diagram generated by Archify. 01 / Agent Execution Lane 02 / Human Governance Lane Upstream (Plan & Design) Midstream (Build & Test) Downstream (Deploy & Maintain) Stage 2: Design · spec.md & plan mode · Agent Execution Lane › Upstream (Plan & Design) Stage 2: Design spec.md & plan mode Stage 3: Build · CLAUDE.md & subagents · Agent Execution Lane › Midstream (Build & Test) · agent speed Stage 3: Build CLAUDE.md & subagents agent speed Stage 4: Test · feedback loop & evals · Agent Execution Lane › Midstream (Build & Test) Stage 4: Test feedback loop & evals Stage 1: Plan · intent.md proto-spec · Human Governance Lane › Upstream (Plan & Design) · approval gate Stage 1: Plan intent.md proto-spec approval gate Stage 5: Deploy · hooks & PR review · Human Governance Lane › Downstream (Deploy & Maintain) · approval gate Stage 5: Deploy hooks & PR review approval gate Stage 6: Maintain · telemetry & incident triage · Human Governance Lane › Downstream (Deploy & Maintain) · closed loop Stage 6: Maintain telemetry & incident triage closed loop approved intent approved plan green evals canary pass breach -> intent.md Legend User UI Agent logic Policy Tool action Cloud service

Continuous Loop Contract

  • • Mỗi stage kết thúc bằng một committed artifact
  • • Human attention tập trung tại các approval gates

Closed-Loop Recovery

  • • Production breach tự động chẩn đoán bởi Agent
  • • Sinh ngược trở lại intent.md để tái khởi động chu trình