亚马逊云科技(AWS)发布了 AWS Emulator,这是一款允许 AI 编码代理在本地模拟 AWS 环境以进行端到端测试的工具 1。该模拟器完全由 AI 代理负责编写和测试,基于覆盖超过 200 种 AWS 服务的 400 多个 API 模型构建,旨在解决代理生成代码后缺乏高效验证环境的问题 1。AWS Emulator 免费且支持离线运行,其代码以 CC0 (Creative Commons Zero) 公共领域贡献许可发布 1。
在具体开发实践中,Matt Goldberg 最初使用 Kiro 工具,仅通过一个提示词便在约 15 分钟内生成了 SQS 模拟器的原型 1。该项目采用了“保真度测试”方法,即基于公开 API 模型设计端到端客户场景,要求代理首先确保测试在真实 AWS 环境中通过,随后据此构建模拟器 1。这一技术路径的背景涉及 DynamoDB 前身 Alf 的开发历史:Swaminathan Sivasubramanian 于 2004 年加入 Amazon,后来架构了 Dynamo,其底层依赖的分布式锁服务 Alf 建于 Paxos 协议之上 1。
随着 AI 代理承担更多编码工作,文章指出工程师的核心职责正发生转变,从传统的代码编写转向设计测试规范、执行保真度检查以及建立自动化安全审查网关 1。在这一新模式下,工程师主要负责定义规范、搭建测试环境并管理合并网关(merge gate),其中包括利用多个安全代理进行对抗性审查 1。
Amazon Web Services has released the AWS Emulator, a tool designed to provide AI coding agents with a local environment for end-to-end testing of AWS services. 1 The emulator is notable for being entirely written and tested by AI agents, relying on over 400 API models that cover more than 200 different AWS services. 1 This development aims to address the critical lack of efficient validation environments for code generated by autonomous agents. 1
The project was spearheaded by Matt Goldberg, who utilized the Kiro agent to generate a prototype for the Simple Queue Service (SQS) emulator in approximately 15 minutes using a single prompt. 1 The resulting software is free to use, operates offline, and has been published under the Creative Commons Zero (CC0) public domain dedication license. 1 To ensure reliability, the team employed "fidelity tests," which are customer-centric end-to-end scenarios derived from public API models; these tests were first validated against real AWS infrastructure before being used to guide the emulator's construction. 1
Historically, this approach reflects broader trends at Amazon regarding foundational infrastructure. Swaminathan Sivasubramanian joined Amazon in 2004 and later architected Dynamo, whose underlying distributed lock service, Alf, was built upon the Paxos protocol. 1 Werner Vogels is also cited among key figures associated with this historical context. 1 The release underscores a significant shift in the engineer's role: rather than writing code directly, engineers now focus on defining specifications, establishing test environments, and implementing merge gates that include adversarial security reviews conducted by multiple AI agents. 1
评论
还没有评论,欢迎留下第一条。