Cisco brings OpenAI’s Codex into the engine room of enterprise software

Cisco has integrated OpenAI’s Codex into its production engineering workflows, using the artificial intelligence system to work across large, interconnected code repositories while meeting enterprise security, compliance and governance requirements.

In a blog post, OpenAI said Cisco worked closely with its engineers to adapt Codex from a developer tool into something that could operate reliably at scale. That meant building in controls for security and compliance, support for long-running tasks and the ability to orchestrate complex workflows rather than perform one-off coding jobs.

For a lay reader, Codex can be thought of as an AI that reads, writes and fixes computer code. What makes Cisco’s deployment notable is not that the AI can write snippets of software, but that it can navigate and reason across thousands of files spread over many repositories, following the same review and approval rules that human engineers must obey.

Cisco described this as “agency”. In practical terms, that means Codex can understand how different parts of a large system fit together, run command-line tools to compile software, test changes and fix errors, and repeat that loop autonomously until the code works. It does this inside Cisco’s existing governance frameworks, rather than bypassing them.

“I’ve loved discovering new opportunities to integrate Codex into Cisco’s enterprise software lifecycle workflows,” said Ching Ho, a member of Cisco’s engineering leadership. He said working with OpenAI to make Codex “enterprise production ready” had focused on adapting it to the realities of large organisations, where changes must be traceable and auditable.

Cisco reported concrete results from the integration. Codex analysed build logs and dependency graphs across more than 15 interconnected repositories, cutting build times by about 20% and saving more than 1,500 engineering hours a month. Build logs record what happens when software is compiled, while dependency graphs map how different components rely on each other. Together, they help identify where delays and failures occur.

Using Codex-CLI, a command-line interface that lets the AI run tools directly, Cisco also automated defect repair in large C and C++ codebases. Those languages underpin much of the world’s networking and systems software and are notoriously complex to maintain. Cisco said this delivered a ten- to fifteen-fold increase in defect resolution throughput, meaning bugs were fixed far faster than before.

Codex was also used to handle framework migrations, the repetitive but risky task of updating large codebases to newer software frameworks. Cisco said work that once took weeks could be compressed into days, with the AI handling repetitive changes autonomously while engineers reviewed the results.

Cisco’s feedback from live production use fed back into OpenAI’s development process. The company said this helped accelerate Codex’s readiness for other large enterprises facing similar challenges.

“Codex has become a meaningful part of how we think about AI-assisted development and operations going forward,” said Brad Murphy, a vice-president leading Cisco’s Splunk engineering team.

The case highlights a shift in how companies are deploying AI in software development. Early tools focused on helping individual programmers write code faster. Cisco’s experience points to a broader role, where AI systems act more like junior engineers, taking on time-consuming maintenance tasks under human supervision.

Related reading

• Microsoft sets out AI-led identity and access security priorities for 2026
• OpenAI turns to age prediction to make ChatGPT safer for teens
• ServiceNow embeds OpenAI models to bring agentic AI into enterprise workflows

For enterprises, the appeal lies less in novelty than in reliability. Large organisations care about security, audit trails and compliance as much as speed. Cisco’s work suggests that AI coding tools are beginning to fit those constraints, moving from experimentation into the core of production systems.

The question for the wider industry is how transferable those gains will be. Cisco’s scale, engineering maturity and close collaboration with OpenAI may not be easy to replicate. But as codebases grow ever larger and more interconnected, the pressure to automate their upkeep is unlikely to ease. Codex’s move into Cisco’s production workflows offers an early indication of how that pressure might be relieved.

The Recap

• Cisco integrated Codex into production engineering workflows at enterprise scale.
• More than 15 repositories and over 1,500 hours saved monthly.
• Cisco gave continuous feedback shaping Codex's enterprise capabilities roadmap.