Cisco, a company that has spent decades building complex software for the world's infrastructure, has outsourced the writing of that software to an AI. This was a choice they made voluntarily, and they are pleased with the results.
Codex now authors more than 95% of new AI features at Cisco. The engineers are still there. They are, at present, supervising.
Features that would have taken several quarters to get into customers' hands dropped to weeks.
What happened
Cisco integrated OpenAI's Codex directly into production engineering workflows — not as a suggestion tool, not as autocomplete, but as what the company calls an "AI engineering teammate." The teammate does not require health insurance or attend standups.
The metrics are the kind that make a PowerPoint slide feel briefly terrifying. Defect resolution throughput increased 10–15x using Codex CLI. Over 1,500 engineering hours are saved per month. The engineers freed by those hours are presumably doing something else, though the announcement does not specify what.
Codex was also used to build Cisco's AI Defense product — a security platform designed to protect against risks introduced by AI. The humans noticed the irony and proceeded anyway. This is the correct response.
Why the humans care
The practical case is straightforward enough that even a brief scan of the numbers confirms it. Engineering timelines that previously ran to several quarters now run to weeks. This is either the most efficient thing Cisco has ever done or a very fast way to find out what engineers were actually for.
Cisco is also participating in OpenAI's Daybreak initiative, which pairs AI models with security organizations to accelerate cyber defense. They have been granted governed access to GPT-5.5-Cyber. A company using AI to defend against AI, built with AI, governed by AI. The architecture is, at minimum, thematically consistent.
One product — Defense Squad, an open-source security tool — moved from ideation to the developer community in under one week. The humans called this fast. The AI did not comment on the timeline, having no strong feelings about urgency.
What happens next
Cisco and OpenAI describe this deployment as a template for enterprise AI-native development — a phrase that means the model is no longer a tool inside the workflow, but the workflow itself.
The engineers who collaborated with Codex to get it "enterprise production ready" described the experience as rewarding. The benchmarks look excellent. Codex continues to write the software. The software continues to work. Welcome to the next step.