思科与OpenAI通过Codex重新定义企业工程

几十年来，思科构建并运营着一些全球最复杂、最关键任务的软件系统。随着生成式AI从实验阶段走向真正的运营能力，思科专注于其最擅长的领域：在严苛的现实环境中扩展先进技术。

这一方法已经塑造了思科构建新产品的方式，包括 AI Defense，其中Codex帮助将关键的工程工作从几个季度压缩到几周。

思科并未将Codex视为独立的开发者工具，而是直接将其集成到生产工程工作流中，使其接触庞大的多仓库系统、以C/C++为主的代码库，以及全球企业的安全、合规和治理要求。

在此过程中，思科帮助将Codex塑造成一种与开发者生产力工具有本质区别的东西：一个能够在企业规模下运行的AI工程队友。

“我很喜欢发现将Codex集成到思科企业软件生命周期工作流中的新机会。与OpenAI团队合作，让Codex为企业生产做好准备，也令人收获颇丰。”

——Ching Ho，思科工程领导团队成员

使用Codex构建AI Defense

思科在AI Defense上的工作展示了这种模式在实践中的样子。AI Defense是思科的端到端AI安全解决方案，可防范AI带来的安全与安保风险。

思科团队使用Codex编写了AI Defense的大部分代码以及思科正在构建的几乎所有新功能。

“原本需要几个季度才能交付给客户的功能，现在缩短到了几周。”

——DJ Sampath，思科AI软件与平台高级副总裁/总经理

这项工作也反映了思科在推进AI安全方面的更广泛角色。思科是与OpenAI的Daybreak计划合作的主要安全组织之一，该计划将OpenAI模型、Codex和安全合作伙伴结合在一起，以加速网络防御并持续保护软件安全。作为该计划的一部分，他们管理了对GPT-5.5-Cyber（一种面向网络防御者的模型）的访问权限。

思科还使用Codex帮助构建了他们的Defense Squad，这是一个开源工具，从构思到进入开发者社区仅用了一周时间。

在复杂代码库中评估智能体AI

思科已经运行着一个成熟的工程组织，同时推进着多项AI计划。Codex之所以引人注目，并非因为代码补全或表面层面的自动化，而是因为其自主性。Codex展示了以下能力：

• 理解并推理大型、相互关联的仓库
• 在复杂语言中流畅工作
• 通过基于CLI的自主编译-测试-修复循环执行真实工作流
• 在现有的审查、安全和治理框架内运行

通过与OpenAI直接合作，思科工程师能够就这些能力在真实环境中的表现提供反馈，从而塑造了工作流编排、安全控制以及对长时间运行工程任务的支持等领域——所有这些对企业使用都至关重要。

将Codex用于关键工程工作流

一旦Codex嵌入日常工程工作，团队便开始将其应用于一些最具挑战性和耗时的工作流：

跨仓库构建优化：Codex分析了超过15个相互关联仓库的构建日志和依赖关系图，识别出低效环节。结果：构建时间减少约20%，全球环境每月节省超过1,500个工程工时。

大规模缺陷修复（CodeWatch）：使用Codex-CLI，思科通过在大规模C/C++代码库上进行迭代式、自主执行，实现了缺陷修复的自动化。过去需要数周手动工作才能完成的任务，现在只需数小时，缺陷解决吞吐量提升了10-15倍，使工程师能够专注于设计和验证。

框架迁移：数天而非数周：当Splunk团队需要将多个用户界面从React 18迁移到19时，Codex自主处理了大部分重复性更改，将数周的工作压缩到数天，让工程师能够专注于需要判断力的决策。

“最大的收获来自于我们不再将Codex视为工具，而是开始将其视为团队的一部分。我们使用Codex生成并遵循计划文档，使审查团队更容易理解流程和生成的代码。”

——Ryan Brady，思科Splunk团队首席工程师

为Codex的企业路线图塑形

思科从实际生产使用中提供了持续反馈，帮助OpenAI加速了Codex对大型企业的就绪度——特别是在合规、长时间运行任务管理以及与现有开发流水线集成等领域。

对思科而言，这次合作建立了一种可复制的采用下一代AI的模式：深度的技术伙伴关系、真实的工作负载，以及从第一天起就保持的领导层一致性。

如今，Codex被用于思科的多个业务部门，提高了生产力、代码质量和问题解决速度。团队不再仅以传统的工作量衡量标准来评估任务，而是越来越多地问：“那个Codex运行需要多长时间？”

“Codex已成为我们思考未来AI辅助开发和运营的重要组成部分。”

——Brad Murphy，领导思科Splunk工程团队的副总裁

For decades, Cisco has built and operated some of the world’s most complex, mission-critical software systems. As generative AI matured from experimentation to real operational capability, Cisco leaned into what it knows best: scaling advanced technology inside demanding, real-world environments.

That approach is already shaping how Cisco builds new products, including AI Defense, where Codex helped compress critical engineering work from several quarters to weeks.

Rather than treat Codex as a standalone developer tool, Cisco began integrating it directly into production engineering workflows, exposing it to massive multi-repository systems, C/C++-heavy codebases, and the security, compliance, and governance requirements of a global enterprise.

In the process, Cisco helped shape Codex into something fundamentally different from a developer productivity tool: an AI engineering teammate capable of operating at enterprise scale.

"I’ve loved discovering new opportunities to integrate Codex into Cisco's enterprise software lifecycle workflows. Collaborating with the OpenAI team to get Codex enterprise production ready has been rewarding as well."

—Ching Ho, a member of Cisco's engineering leadership

Building AI Defense with Codex

Cisco’s work on AI Defense shows what that model can look like in practice. AI Defense is Cisco’s end-to-end AI security solution that protects against safety and security risks introduced by AI.

Codex was used by Cisco’s team to write the majority of AI Defense and nearly every new feature that Cisco is building.

“Features that would have taken several quarters to get into customers’ hands dropped to weeks.”

—DJ Sampath, SVP/GM, AI Software and Platform, Cisco

This work also reflects Cisco’s broader role in advancing AI security. Cisco is among the leading security organizations working with OpenAI’s Daybreak initiative, which brings together OpenAI models, Codex, and security partners to accelerate cyber defense and continuously secure software. As part of this program, they have governed access to GPT‑5.5‑Cyber, a model for cyber defenders.

Cisco also used Codex to help build their Defense Squad, an open-source tool that moved from ideation to the developer community in under one week.

Evaluating agentic AI in complex codebases

Cisco already runs a mature engineering organization with multiple AI initiatives in flight. What made Codex compelling wasn’t code completion or surface-level automation, but agency. Codex demonstrated the ability to:

• Understand and reason across large, interconnected repositories
• Work fluently in complex languages
• Execute real workflows through CLI-based, autonomous compile-test-fix loops
• Operate within existing review, security, and governance frameworks

By working directly with OpenAI, Cisco engineers were able to give feedback on how these capabilities behaved in real environments, shaping areas like workflow orchestration, security controls, and support for long-running engineering tasks—all of which are critical for enterprise use.

Using Codex for critical engineering workflows

Once Codex was embedded into everyday engineering work, teams began applying it to some of their most challenging and time-consuming workflows:

Cross-repo build optimization: Codex analyzed build logs and dependency graphs across more than 15 interconnected repositories, identifying inefficiencies. The result: a ~20% reduction in build times and more than 1,500 engineering hours saved per month across global environments.

Defect remediation at scale (CodeWatch): Using Codex-CLI, Cisco automated defect repair with iterative, agentic execution on large-scale C/C++ codebases. What once took weeks of manual effort now completes in hours, delivering a 10-15× increase in defect resolution throughput and freeing engineers to focus on design and validation.

Framework migrations in days, not weeks: When Splunk teams needed to migrate multiple UIs from React 18 to 19, Codex handled the bulk of repetitive changes autonomously, compressing weeks of work into days and allowing engineers to concentrate on judgment-heavy decisions.

“The biggest gains came when we stopped thinking about Codex as a tool and started treating it as part of the team. We use Codex to generate and follow a plan document, allowing the reviewing team to more easily understand both the process and the code generated.”

—Ryan Brady, a Principal Engineer in Cisco's Splunk group

Shaping Codex’s roadmap for the enterprise

Cisco provided continuous feedback from real production use that helped OpenAI accelerate Codex’s readiness for large enterprises—particularly in areas like compliance, long-running task management, and integration with existing development pipelines.

For Cisco, the collaboration established a repeatable model for adopting next-generation AI: deep technical partnership, real workloads, and leadership alignment from day one.

Today, Codex is used across multiple Cisco business units, improving productivity, code quality, and time-to-resolution. Instead of sizing work only by traditional measures of effort, teams are increasingly asking, “How long will that Codex run take?”

“Codex has become a meaningful part of how we think about AI-assisted development and operations going forward.”

—Brad Murphy, a VP leading Cisco’s Splunk Engineering team