Codex can control your Mac

OpenAI has expanded Codex from a coding agent into a macOS tool that can operate apps on a user’s screen as part of a workflow. (openai.com) OpenAI said the updated Codex app for macOS adds “computer use,” which lets the model see graphical interfaces after users install a plugin and grant Screen Recording and Accessibility permissions in macOS. At launch, OpenAI’s developer docs say the feature is available on macOS, but not in the European Economic Area, the United Kingdom, or Switzerland. (developers.openai.com) The same release adds image generation inside Codex through GPT-Image-1.5, OpenAI’s image model, so users can generate and revise visuals without leaving the app. OpenAI said the desktop app also adds an in-app browser, memory, plugins, and support for reviewing pull requests, opening multiple files and terminals, and connecting to remote development boxes over Secure Shell. (openai.com) Computer-use tools turn an artificial intelligence model into something closer to a software operator: it reads pixels on a screen, then moves a cursor, clicks buttons, and types into apps. OpenAI is applying that approach to Codex, which it introduced in 2025 as a cloud-based software engineering agent for parallel coding tasks inside ChatGPT. (developers.openai.com) (openai.com) That puts Codex closer to the broader “agent” model that artificial intelligence companies have been chasing over the past year: software that does multi-step work across tools instead of only answering prompts in a chat box. OpenAI’s April 16, 2026 product post says Codex can now “operate your computer alongside you,” learn from previous actions, and take on repeatable work. (openai.com) The image piece matters for software teams because interface work often requires code and assets in the same loop. OpenAI said GPT-Image-1.5 is already available in its image stack and described it in December 2025 as a faster model for generating and editing images with more precise edits and more consistent details. (openai.com) OpenAI first launched the Codex desktop app for macOS in February 2026 as what it called a “command center for agents,” with separate project threads and parallel workflows. The new release pushes that app beyond code generation by letting it act inside desktop software and web previews while a user keeps working. (openai.com 1) (openai.com 2) OpenAI has framed the update as part of a larger expansion of Codex into long-running engineering work. Its current Codex product page says the app is built for planning, building features, refactors, reviews, and releases with multiple agents working across projects. (openai.com) The practical test now is whether developers trust an artificial intelligence agent with the same permissions they would give a human assistant on a Mac. OpenAI’s rollout keeps a human in the permission loop, but the product is now aimed at a much wider slice of day-to-day computer work than coding alone. (developers.openai.com) (openai.com)