The Codex app represents OpenAI’s clearest move yet toward an agent-native future, where developers don’t just write code with AI—they direct, supervise, and collaborate with entire teams of autonomous agents across real software projects. (Source: Image by RR)

OpenAI Demonstrates Codex’s Power in Building, Testing a Complete 3D Game

OpenAI has introduced the Codex app for macOS, a new desktop command center designed to manage multiple AI coding agents simultaneously, marking a significant shift in how software is built and maintained. Rather than pairing with a single assistant for short tasks, developers can now supervise coordinated teams of agents working in parallel across long-running projects. The app reflects a broader evolution in AI development, where the challenge is no longer what agents can do, but how humans can effectively direct and collaborate with them at scale.

At its core, the Codex app allows developers to run agents in isolated threads organized by project, enabling seamless multitasking without losing context. Built-in support for Git worktrees ensures that multiple agents can safely operate on the same repository without conflicts, while developers retain the ability to review diffs, comment on changes, or manually intervene in their editor. The app, as noted at openai.com, syncs with existing Codex CLI and IDE workflows, allowing users to pick up exactly where they left off across environments.

Beyond coding, Codex now supports Skills, modular bundles of instructions, tools, and scripts that let agents perform broader knowledge work such as research, documentation, deployment, design translation, and image generation. OpenAI demonstrated this capability by having Codex independently build and test a full 3D racing game using more than seven million tokens, assuming roles ranging from designer to QA tester. These skills can be shared across teams and reused anywhere Codex operates, transforming the agent from a code generator into a general-purpose work executor.

The app also introduces Automations, enabling Codex to run scheduled background tasks like issue triage, CI failure summaries, and release reporting. Security remains a central focus, with sandboxed permissions, configurable execution rules, and explicit approval required for elevated actions. Available today on macOS and included with ChatGPT subscriptions—along with temporarily expanded access for Free and Go users—the Codex app signals OpenAI’s push toward agent-native software development, with Windows support and deeper automation planned next.

read more at openai.com