OpenAI is extending its full-duplex voice capabilities to developer tools, allowing users to control Codex and ChatGPT Work on desktop systems using natural speech.
Two weeks after debuting the GPT-Live model, which can listen and speak simultaneously, OpenAI is integrating this technology into ChatGPT desktop applications for macOS and Windows. The update connects the conversational voice layer directly with agentic systems like Codex and ChatGPT Work. The feature allows software engineers to manage multi-threaded coding jobs, review pull requests, and debug applications without touching a keyboard.
According to an OpenAI spokesperson, this marks the first time voice activation has been added to these specific desktop experiences. The company released a video showing developers Jason Liu and Guinness Chen using the feature in the same room, issuing different instructions to the same model session. This setup demonstrates how the system manages multiple voices and distinct commands within a single interaction.
Related: Enterprise AI agent governance faces significant gaps
The underlying architecture separates the real-time voice layer from the execution engines. While GPT-Live maintains a fluid conversation—inserting natural acknowledgments like “got it” without interrupting the user—it offloads heavy computational workloads to background reasoning models. On macOS, the desktop app incorporates “Appshots” and screen context features, enabling the system to analyze the frontmost window alongside local files and codebase structures.
Coordinating complex tasks with voice
This decoupling creates a dynamic where developers talk through problems conversationally while agents execute tasks asynchronously. Rather than stopping to type detailed instructions or switch windows, users can direct the system hands-free. The full-duplex engine decides when to speak, pause, or invoke tools, maintaining conversational state as background agents process code modifications.
The central operational capability enables multi-task execution across Codex and ChatGPT Work environments. A developer can initiate multiple concurrent task threads from a single prompt, such as investigating an authentication bug, reviewing a pending pull request, and generating missing unit tests simultaneously. The desktop application coordinates these actions across disparate contexts, tracing issues through Slack conversations, GitHub repositories, and local codebases.
Related: Data Modernization Fails When No One Understands It
Developers can also verbally convert design mockups into working code, splitting tasks across frontend, backend, and testing layers. With support for multi-folder projects and remote execution via iOS, engineers can check task progress and redirect active jobs without switching applications. This approach shifts the workflow from managing individual processes line by line to overseeing broader project orchestration.
OpenAI’s voice-enabled desktop release operates under a proprietary, commercial enterprise model. Access is restricted to paid subscribers across Plus, Pro, Business, Enterprise, and Education plans.
