Key facts
- OpenAI introduced reusable cloud development environments for its Codex software engineering agent.
- The new environments allow access from any device, including phones and remote cloud instances.
- Codex now supports voice commands through an updated CLI.
- A new code review experience integrates Codex into the ChatGPT desktop app.
- Codex Security Cloud tools are available to scan GitHub repositories and prepare fixes.
- OpenAI also announced a new Decisions API and an updated Agents API supporting computer use.
OpenAI has enhanced its software engineering agent, Codex, with new capabilities aimed at making it more versatile and accessible. The update, announced at OpenAI's Dev Day, includes persistent and reusable cloud development environments that can be accessed from any device, moving beyond isolated remote sandboxes.
These new environments are designed to speed up task initiation and provide teams with a shared workspace featuring approved settings and permissions. Developers can now run Codex on their computers, remotely from their phones, or directly in the cloud.
Further enhancements include an updated Codex CLI that allows developers to start and direct tasks using voice commands. A new '/agents' view is intended to simplify delegating work and tracking multiple tasks concurrently. Improvements have also been made to everyday tasks like editing prompts, resuming sessions, and using worktrees, alongside a redesigned terminal user interface for better readability during long sessions.
Codex is also being integrated into a new code review experience within the ChatGPT desktop app. This feature allows users to read summaries, examine code changes, and inquire about potential issues before submitting feedback on platforms like GitHub and GitLab. Automatic code reviews are also available, enabling Codex to perform initial checks while users are away.
On the security front, OpenAI is launching Codex Security Cloud, a suite of tools designed to help developers secure their infrastructure. These tools can scan GitHub repositories on demand or on a schedule, and investigate findings, remove duplicates, and prepare fixes in the cloud, even when a user's laptop is inactive. This toolset also provides access to models from OpenAI's Daybreak Blue initiative without requiring a separate application.
Additionally, OpenAI announced several API updates, including a new Decisions API for real-time decision-making using Luna to provide predefined answers to user-defined questions. The Agents API has been updated to support computer use and integration with Amazon's Bedrock Managed Agents for building OpenAI agents that run entirely on AWS.