OpenAI launches Codex: the AI agent that writes code instead of engineers

OpenAI has released a cinematic video "It's time to fly" — a signal that Codex, their flagship AI agent for development, has moved from beta-tool mode to a full-fledged product.

What Codex can do in 2026

Codex is built on GPT-5.5 — OpenAI's first fully retrained base model since the release of GPT-4.5. The model was trained with agent scenarios in mind: it verifies its own code before delivering results and can conduct hundreds of sequential tool calls without human intervention. The key innovation is parallel execution in multiple sandbox environments simultaneously:

• runs a test suite in one environment
• generates documentation from diffs in parallel
• proposes refactoring in a third sandbox
• reads the screen and controls GUI applications for QA automation
• integrates into CLI, IDE, ChatGPT, and GitHub within a single account

While an engineer reviews the results, Codex is already working on the next task.

Numbers and scale

The product launch did not go unnoticed. At the time the promo was released, Codex had over 4 million active developers per week. NVIDIA provided access to the agent for 10,000 of its employees — not only engineers but also non-technical teams. Harness, a DevOps company, published a case study on OpenAI's website about how Codex is transitioning from assistant to full-fledged member of the engineering process — an "agent-first world" where an agent completes the task rather than suggesting how to complete it.

What it means

The "It's time to fly" slogan precisely describes the shift: Codex stops being smart autocomplete and becomes an autonomous executor. For development teams this means that routine engineering tasks — tests, documentation, small refactorings — are increasingly delegated to the agent, while people focus on architectural decisions and code review.