Cloudflare 10x Scaling · Azure AI Sandboxes

一、 权威必看

EN: Cloudflare has successfully optimized its Security Insights system to process over 120 scans per second globally. By refining Kafka consumers, Postgres queries, and API structures, the company achieved a tenfold increase in scanning capacity without adding any new hardware infrastructure. This technical milestone demonstrates how software-level architectural optimization can drastically enhance security monitoring capabilities for all customers.

中： Cloudflare 对其安全洞察系统进行了深度优化，目前在全球范围内每秒可处理超过 120 次扫描。通过重构 Kafka 消费者、Postgres 查询逻辑以及 API 架构，该公司在不增加任何新硬件基础设施的情况下，实现了扫描容量的十倍增长。这一技术里程碑展示了软件层面的架构优化如何显著提升对所有客户的安全监控能力。这种无需额外硬件投入即可实现性能飞跃的模式，为大型分布式系统的弹性扩展提供了极具参考价值的工程实践范例，表明在存量资源下通过代码级调优挖掘潜力，是应对海量数据挑战的有效路径。

二、 深度与多元

EN: The Angular team at Google has released the angular/skills repository, which focuses on Agent Skills to enhance AI coding agents’ ability to write modern Angular code. This repository includes skills for generating code and scaffolding applications, reinforcing current Angular conventions. It serves as a snapshot, aiming to improve AI suggestions by providing updated context, thereby bridging the gap between traditional development workflows and AI-assisted programming.

中： Google 的 Angular 团队发布了 angular/skills 仓库，专注于 Agent Skills 以增强 AI 编程代理编写现代 Angular 代码的能力。该仓库包含用于生成代码和搭建应用程序的技能，强化了当前的 Angular 规范。它作为一个快照，旨在通过提供更新的上下文来改善 AI 建议的质量，从而弥合传统开发工作流与 AI 辅助编程之间的差距。这一举措标志着前端框架官方开始主动介入 AI 生态建设，通过标准化技能定义，确保 AI 生成的代码符合最佳实践，减少因模型幻觉导致的架构偏差，为开发者提供了更可控、更精准的 AI 协作工具。

三、 科技与财经

EN: Microsoft has announced the public preview of Azure Container Apps Sandboxes, a new ARM resource type (Microsoft.App/SandboxGroups) designed to run untrusted code generated by agents in hardware-isolated environments. Each sandbox starts from an OCI disk image in less than a second and can scale to thousands of instances simultaneously while costing nothing when idle. This innovation addresses critical security concerns in AI agent deployment by ensuring that potentially malicious or unstable code cannot compromise the host system.

中： 微软宣布了 Azure Container Apps Sandboxes 的公开预览版，这是一种新的 ARM 资源类型（Microsoft.App/SandboxGroups），旨在在硬件隔离的环境中运行由代理生成的不可信代码。每个沙箱从 OCI 磁盘镜像启动仅需不到一秒的时间，可同时扩展至数千个实例，且在空闲时零成本。这一创新通过确保潜在恶意或不稳定的代码无法损害主机系统，解决了 AI 代理部署中的关键安全问题。对于企业而言，这意味着可以在云端安全地执行由 LLM 生成的任意代码片段，无需担心资源耗尽或权限泄露，极大地降低了 AI Agent 在企业内部落地的技术门槛和安全风险。

四、 国际视野

EN: David Stein shared insights on rethinking large-scale architectural migrations using AI, highlighting ServiceTitan’s “assembly line” pattern. He explained how decomposing legacy codebase refactoring into standardized tasks can achieve massive parallelization. The presentation emphasized the critical role of programmatically rigid validation loops to eliminate LLM hallucinations and accelerate engineering agility, demonstrating a practical approach to migrating complex systems in weeks rather than years.

中： David Stein 分享了利用 AI 重新思考大规模架构迁移的见解，重点介绍了 ServiceTitan 的“流水线”模式。他解释了如何将遗留代码库的重构分解为标准化的任务，以实现大规模的并行处理。演讲强调了编程严格的验证循环在消除 LLM 幻觉和加速工程敏捷性方面的关键作用，展示了在数周内而非数年内迁移复杂系统的实用方法。这种将非结构化的重构工作转化为结构化、可验证的流水线作业的思路，为传统企业应对技术债务提供了清晰的执行路径，证明了 AI 不仅是代码生成工具，更是流程再造的核心驱动力。

五、 青年与生活

EN: At the Fourth Software Innovation Development Conference in Wuhan, it was announced that the city’s software industry scale exceeded 400 billion yuan in 2025, ranking in the top ten nationally. Looking ahead to the end of the “15th Five-Year Plan,” Wuhan aims to break through 800 billion yuan. The core driver for this doubling is the deep integration of AI and software, with a focus on industrial manufacturing as the main battlefield for physical AI value realization.

中： 在武汉第四届软件创新发展大会上获悉，2025年武汉市软件产业规模首次突破4000亿元，跻身全国城市前十。面向“十五五”期末，武汉力争突破8000亿元的奋斗目标。这一翻番增长空间的核心动力正来自AI与软件的深度融合，工业制造被视为物理AI价值落地的主战场。武汉通过集聚22家软件上市公司及近百家头部企业代表，构建“5个综合集聚区+9个特色集聚区”的空间布局，旨在从“软件大省”向“软件强省”跃迁。这一产业动向不仅反映了区域经济的数字化转型决心，也为相关领域的从业者提供了明确的职业发展和创业机遇方向。

【21ZHAO 综合判断】

EN: The convergence of these five dimensions reveals a clear trend: AI is transitioning from a generative tool to an operational infrastructure component. Cloudflare’s optimization shows how existing systems can be scaled without hardware costs, while Microsoft’s sandboxes provide the security framework for executing untrusted AI code. Angular’s skills and ServiceTitan’s migration patterns demonstrate that standardizing AI interactions is key to enterprise adoption. Developers should focus on building robust validation layers and secure execution environments.

• Implement strict input/output validation loops when integrating LLMs into critical workflows to prevent hallucinations from affecting production systems.
• Leverage hardware-isolated sandbox environments for any code generation tasks involving untrusted or third-party AI models to mitigate security risks.

中： 这五个维度的事件共同揭示了一个清晰趋势：AI 正从生成式工具转变为运营基础设施的核心组件。Cloudflare 的优化展示了如何在无硬件成本的情况下扩展现有系统，而微软的沙箱则为执行不可信 AI 代码提供了安全框架。Angular 的技能包和 ServiceTitan 的迁移模式表明，标准化 AI 交互是企业落地的关键。开发者应专注于构建强大的验证层和安全执行环境，以应对日益复杂的 AI 集成需求。

• 在将 LLM 集成到关键工作流时，实施严格的输入/输出验证循环，防止幻觉影响生产系统稳定性。
• 对于涉及不可信或第三方 AI 模型的代码生成任务，利用硬件隔离的沙箱环境以降低安全风险。

参考来源

• [权威要闻]：Scaling Security Insights: how we achieved a 10x increase in global scanning capacity - 原文链接
• [深度解读]：Angular’s Official Agent Skills Helps AI Coding Tools Write Modern Angular - 原文链接
• [科技财经]：Run Untrusted AI Agent Code Safely with Azure Container Apps Sandboxes - 原文链接
• [国际视野]：Presentation: Moving Mountains: Migrating Legacy Code in Weeks instead of Years - 原文链接
• [青年声音]：武汉软件产业“瞄准”8000亿 让AI从“动口”走向“动手” - 原文链接