Anthropic 官方 · 工程方法
AI 原生软件开发生命周期
The AI-Native SDLC Playbook
从意图、规范、计划、构建、持续评估到生产反馈，建立有人类判断、有自动化证据的 AI 原生交付闭环。
课程目录
- 01 代码不再是瓶颈Code Is No Longer the Bottleneck 对应官方章节：Code is no longer the bottleneck
- 02 把线性流程改造成证据闭环The AI-Native Loop 对应官方章节：What is an AI-native SDLC?; Plays
- 03 Plan：把原始想法固化为 intent.mdPlan: Capture Intent 对应官方章节：Stage 1 — Plan; Capture as intent.md
- 04 Design：用 spec.md 与 Skills 前置约束Design: Specs and Skills 对应官方章节：Stage 2 — Design; Skills as institutional knowledge
- 05 Build：先计划，再让上下文服务实现Build: Plan and Context 对应官方章节：Stage 3 — Build; plan mode; CLAUDE.md
- 06 Test：把验证与持续评估织进实现Test: Continuous Evals 对应官方章节：Stage 4 — Test; Continuous evals in CI
- 07 Deploy：分层评审、门禁与 CI/CDDeploy: Review and CI/CD 对应官方章节：Stage 5 — Deploy
- 08 Maintain：从控制带异常回到新意图Maintain: Close the Feedback Loop 对应官方章节：Stage 6 — Maintain; Closing the loop
- 09 治理：控制、证据、责任与度量Governance and Measurement 对应官方章节：Governance considerations; How to measure it
- 10 落地顺序与成熟度自评Adoption Checklist and Maturity Assessment 对应官方章节：Plays; Closing thoughts; Resources and acknowledgments