OpenAI's Codex Update Adds macOS App Control
OpenAI has enhanced its Codex agentic coding and development system with several new capabilities. The most notable addition allows Codex to operate desktop applications directly on users' computers. According to OpenAI's blog post, Codex handles these tasks in the background without disrupting other work. Multiple agents can run simultaneously. Developers benefit from this for testing frontend modifications, app trials, or interactions with software lacking APIs.
An example shows Codex playing Tic Tac Toe on a desktop. The rollout begins today for Codex desktop app users logged in via ChatGPT. Access starts on macOS only. OpenAI provided no schedule for Windows or Linux support. Users in the European Union face a delay, with availability promised soon.
Expanded Capabilities in Codex
Codex now generates and refines images using gpt-image-1.5. It integrates new plugins for GitLab, Atlassian Rovo, and the Microsoft Suite. Native web browsing arrives through an in-app browser. Users can add comments on web pages to give exact directions to the agent.
Task automation improves too. People can reuse prior conversation threads. Codex schedules its own future jobs and activates automatically to resume long-running projects.
A memory function joins the lineup. It stores context from previous sessions, like user preferences, fixes, and hard-won details. This opt-in preview aims to speed up tasks and boost quality without needing lengthy setup each time. Rollout hits Enterprise, Education, and EU accounts shortly.
Competition in AI Coding Tools
Stay ahead of the AI curve
The most important updates, news, and content — delivered weekly.
No spam. Unsubscribe anytime.
These changes position OpenAI against rivals like Google and Anthropic. The focus on Anthropic stands out after Claude Code's strong performance. OpenAI pushes to match that progress in automated coding and software work.
Codex builds on OpenAI's line of developer tools. The company, started in 2015, leads in generative AI with products like ChatGPT. Anthropic, founded by former OpenAI staff in 2021, gained attention with Claude models known for safety and coding prowess. Claude Code has drawn praise for handling complex programming tasks autonomously.
Robert Hart, the London-based AI reporter who covered this for The Verge, notes OpenAI's quick response to competitors. His prior work at Forbes included health, science, and tech topics. This update reflects broader trends where AI agents take over routine development chores.
OpenAI plans further expansions, though specifics on non-macOS platforms remain unclear. The memory preview and EU access signal ongoing refinements. Developers already test the desktop control on Macs, reporting gains in efficiency for app interactions without code tweaks.
As AI coding assistants evolve, features like background operation and persistent memory set Codex apart. Parallel agents enable batch testing, a boon for teams. Image iteration with gpt-image-1.5 supports UI design workflows. Plugins connect to popular dev tools, easing integrations.
The in-app browser with annotations refines web-based instructions, reducing errors in agent actions. Scheduling and auto-wake handle multi-day projects, vital for real-world software cycles.