Rakuten fixes issues twice as fast with Codex

Rakuten 是一家在电商、金融科技和移动通信等领域开展业务的全球性创新公司，面向消费者和商家提供大规模服务。公司在全球拥有 30,000 名员工，其工程团队在庞大且复杂的产品生态中交付产品，既要追求速度，也要保证可靠性。

因此， Rakuten 负责商务 AI 的总经理 Yusuke Kaji 在过去一年里，推动将具备“代理能力”的工作流更深入地嵌入团队的规划、开发和验证环节。来自 OpenAI 的编码代理 Codex 已成为 Rakuten 工程栈的核心部分，尤其在公司需要提速但又不能牺牲安全性的场景中，发挥了重要作用。

在过去一年中， Rakuten 的工程师在运维和软件交付环节广泛使用 Codex，压缩了故障响应时间（平均恢复时间即 MTTR 下降约 50%）、通过自动化代码审查与漏洞检测强化 CI/CD，支持在复杂项目上更高程度的自主开发。

“我们关心的不只是快速生成代码，” Kaji 说，“我们关心的是安全交付。没有安全的速度不是成功。”

主要成效与优先方向

在 Rakuten 的工程体系里，AI 议程清晰且以落地为导向。 Kaji 将工作围绕三项优先目标组织，团队围绕这三点展开实践：

• 更快交付（“Speed!! Speed!! Speed!!”）：团队在运维工作流中调用 Codex，结合基于 KQL 的监控与诊断，加速根因分析与修复，使 MTTR 最多压缩约 50%。
• 更安全交付（“Get things done”）：在 CI/CD 中引入 Codex 进行代码审查与漏洞检测，把公司的编码规范自动化地应用到流程里，从而在有护栏的情况下加速发布。
• 更智能运维（“ AI-nization ”）：在更大、更模糊的项目上， Codex 可推动工作从需求走向可运行的实现，减少对完备需求的依赖，提升自治执行力，将原本需要一个季度的工作压缩到几周。

作为更大工具包中的可靠代理， Codex 在需要速度、安全与自治产生复合价值的环节中发挥直接作用。

加快响应、压缩故障恢复时间

在 Rakuten，“速度”不仅指开发节奏，还包括恢复时间。团队使用 KQL（ Azure 的日志与遥测查询系统）监控 API 并分析信号， Codex 与这些工作流并行，帮助识别根因并建议修复方案，从告警到解决的时间显著缩短。

从 SRE 视角看，这缩短了从检测到修复的路径。工程师不再需要手工拼接查询、日志和补丁，而是可以把精力放在验证与部署修复上。 Rakuten 估计，这种做法在问题发生时可以将 MTTR 缩短约 50%。换句话说：出现故障时，借助 Codex 的修复速度提升了约一倍。

在 CI/CD 中调用 Codex 以提高安全性

当交付加速时，审查与部署容易成为瓶颈。 Rakuten 的做法是将 Codex 直接嵌入其 CI/CD 管道中，在变更进入生产前执行代码审查与漏洞检测。公司把内部的编码原则与标准输入到这些工作流里，让自动审查结果与公司期望保持一致。

“我们把内部的编码原则提供给 Codex，” Kaji 说，“它用同样的原则来判断代码是否符合我们的标准。”

结果是：安全检查以一致且自动化的方式发生，使团队能在不降低标准的前提下更快推进。

以单一规格驱动全栈构建，提升自治能力

第三项优先目标“ AI-nization ”侧重于提升自治。 Codex 不仅用于审查和维护，还能端到端执行更大、定义不完全的项目。从部分需求出发， Codex 能推进开发并产出可用成果。

例如，将已有的基于 Web 的 AI 代理服务做成移动应用。 Codex 在没有逐步人工指导的情况下，完整实现了规格：构建了包含 Python/FastAPI 后端和 Swift/SwiftUI iOS 客户端的全栈实现，并实现所需的后端 API。该项目的开发周期从一个季度被压缩到数周。

工程师角色从“编写”转为“验证”

随着 Codex 承担越来越多的代码生成工作， Rakuten 将工程师的职责转向撰写更清晰的规格并按可衡量的标准验证输出。

“我们的角色不再是检查每一行代码，” Kaji 说，“而是清晰定义我们想要的成果，并建立验证方法。”

为支撑这一转变， Rakuten 在工程、产品与非技术团队中开展了实操型工作坊，使 Codex 成为帮助团队更快交付、更安全运营、并在组织范围内扩展自治开发能力的核心工具。

加入新工作时代

全球已有超过 100 万家企业借助 OpenAI 取得切实成果。欲了解更多，请联系销售。

Rakuten⁠ is a global innovation company operating across e-commerce, fintech, and mobile communications, serving both consumers and merchants at massive scale. With 30,000 employees worldwide, its engineering teams ship across a large, complex product ecosystem where both speed and reliability are essential.

That’s why Yusuke Kaji, General Manager of AI for Business at Rakuten, has spent the past year pushing agentic workflows deeper into how teams plan, build, and validate software. Codex⁠—the coding agent from OpenAI—has become a core part of Rakuten’s engineering stack, especially where the company needs to move faster without compromising security.

Over the past year, Rakuten engineers have used Codex across operations and software delivery to compress incident response (including a ~50% reduction in mean time to recovery, or MTTR), strengthen CI/CD with automated code review and vulnerability checks, and support more autonomous development on complex projects.

“We don’t just care about generating code quickly,” Kaji says. “We care about shipping safely. Speed without safety is not success.”

50% faster recovery and quarters-to-weeks ship cycles

Inside Rakuten’s engineering team, their AI agenda is crisp and intentionally operational. Kaji frames the work around three priorities that teams rally behind:

• Build faster (“Speed!! Speed!! Speed!!”): Teams use Codex in operational workflows, including KQL-based monitoring and diagnosis, to accelerate root-cause analysis and remediation, helping compress MTTR by up to 50%.
• Build safer (“Get things done”): Codex is invoked in CI/CD for code review and vulnerability checks, applying internal standards automatically so teams can ship quickly with guardrails.
• Operate smarter (“AI-nization”): Codex drives larger, ambiguous projects forward from specification toward working implementations, reducing dependence on perfectly-defined requirements, enabling more autonomous execution, and ultimately compressing quarter-long efforts into weeks.

Codex maps directly to each priority as a dependable agent in a broader toolkit, showing up where speed, safety, and autonomy create compounding value.

Building faster by compressing incident response

Speed at Rakuten includes recovery time, not just development velocity.

Teams use KQL (Azure’s query system for logs and telemetry) to monitor APIs and analyze signals. Codex works alongside these workflows to help identify root causes and suggest fixes, reducing the time between alert and resolution.

From a site reliability engineering (SRE) perspective, this shortens the path from detection to remediation. Instead of manually stitching together queries, logs, and patches, engineers can focus on validating and deploying fixes.

Rakuten estimates this approach can reduce MTTR by approximately 50% when issues occur. Or more simply put: Rakuten has used Codex to fix problems twice as fast when something breaks.

Building safer by invoking Codex in CI/CD

As shipping accelerates, review and deployment can become bottlenecks. Rakuten addresses this by integrating Codex directly in its CI/CD pipeline.

Codex conducts code review and vulnerability checks before changes reach production. Rakuten feeds internal coding principles and standards into these workflows so reviews align with company expectations.

“We provide our internal coding principles to Codex,” Kaji says. “Using the same principles, it reviews whether the code aligns with our standards.”

The result: safety checks happen consistently and automatically, enabling teams to move faster without lowering standards.

Build smarter by executing full-stack builds from a single spec

Rakuten’s third priority—AI-nization—focuses on autonomy. Codex is used not only for review and maintenance, but also for executing larger, ambiguous projects end-to-end. Instead of requiring perfectly defined specifications, Codex can move forward from partial requirements and produce usable artifacts.

“The latest Codex models can read between the lines,” Kaji says. “Even if the requirements are not perfectly defined, it understands what we’re trying to build.”

One example: building a mobile app version of an existing web-based AI agent service. Codex implemented the entire specification, involving a full stack implementation with a Python/FastAPI backend and a Swift/SwiftUI iOS app, including all the backend APIs, without step-by-step human instruction. Codex cut the development time for this project from one quarter to weeks.

Shifting engineering from writing to verifying

As Codex takes on more code generation work, Rakuten is shifting the engineer’s role to writing clearer specifications and verifying outputs against measurable standards. 

“Our role is not to check every line of code anymore,” Kaji says. “Our role is to define clearly what we want and establish how to verify it.”

Rakuten has supported this shift through hands-on workshops across engineering, product, and non-technical teams—contributing to Codex playing a central role in helping teams ship faster, operate more safely, and scale autonomous development across the organization.

Join the new era of work

Generated by RSStT. The copyright belongs to the original author.

Source