OpenAI's Codex Update Lets Agents Click, Type and Schedule Work

OpenAI has released a major update to Codex that lets the AI agent control a user's computer directly — seeing, clicking, and typing with its own cursor — as the company pushes the tool beyond writing code and into the full software development lifecycle.

The update, rolling out now to Codex desktop app users signed in with ChatGPT, targets the more than 3 million developers OpenAI says use Codex every week. Alongside computer use, the release adds an in-app browser, image generation with gpt-image-1.5, more than 90 new plugins, expanded automations that can run for days or weeks, and a preview of memory that lets the agent recall preferences and corrections from previous sessions.

The stakes are competitive. Coding agents have become the fastest-growing category of AI products, and every major vendor — Anthropic, Google, and OpenAI itself — is racing to expand them from code generation into autonomous, general-purpose work assistants. OpenAI's bet is that developers will adopt agents that consolidate tasks currently scattered across editors, terminals, browsers, project management tools, and chat apps.

Agents that share your screen

The headline feature is what OpenAI calls background computer use. Codex can now operate all the apps on a user's computer by seeing, clicking, and typing with its own cursor. Multiple agents can run in parallel on a Mac without interfering with the user's own work in other apps. OpenAI positions this as useful for iterating on frontend changes, testing apps, or working in applications that don't expose an API — a common limitation for agents that rely on formal integrations.

Computer use is initially available on macOS only, with a rollout to EU and UK users promised soon. That geographic delay reflects ongoing European privacy and compliance constraints that have shaped OpenAI's feature rollouts across its product line.

Codex is also beginning to work natively with the web. The app now includes an in-app browser where users can comment directly on pages to give the agent precise instructions. OpenAI says this is useful for frontend and game development today, and the company plans to expand it over time so Codex can fully command the browser beyond web applications running on localhost.

Image generation rounds out the multimodal additions. Codex can now use gpt-image-1.5 to generate and iterate on images, which OpenAI says helps when creating visuals for product concepts, frontend designs, mockups, and games — combined with screenshots and code inside the same workflow.

Ninety new plugins

The release adds more than 90 plugins that combine skills, app integrations, and MCP servers, giving Codex more ways to gather context and take action across a developer's tools. Named plugins include Atlassian Rovo for managing JIRA, CircleCI, CodeRabbit, GitLab Issues, Microsoft Suite, Neon by Databricks, Remotion, Render, and Superpowers.

The MCP support matters beyond OpenAI's ecosystem. The Model Context Protocol has emerged as a de facto standard for connecting AI agents to external tools, and Codex's deepening plugin architecture signals OpenAI's intent to make its agent a hub for the growing library of MCP-compatible services.

Built for the whole development lifecycle

The Codex app itself has gained features aimed at real developer workflows: support for addressing GitHub review comments, multiple terminal tabs, and — in alpha — connecting to remote devboxes over SSH. Users can now open files directly in the sidebar with rich previews for PDFs, spreadsheets, slides, and documents, and a new summary pane tracks agent plans, sources, and artifacts.

OpenAI frames these additions as a way to collapse the cycle of writing code, checking outputs, reviewing changes, and collaborating with the agent into one workspace, rather than forcing developers to bounce between an editor, a browser, and a terminal.

Work that persists across days

The most consequential changes may be the ones that extend Codex across time. OpenAI has expanded automations so that existing conversation threads can be reused, preserving previously built-up context. Codex can now schedule future work for itself and wake up automatically to continue long-term tasks — potentially across days or weeks.

Teams already use automations for tasks like landing open pull requests, following up on assignments, and keeping up with fast-moving conversations across Slack, Gmail, and Notion, according to OpenAI.

A preview of memory takes persistence further. Codex can now remember useful context from previous experience, including personal preferences, corrections, and hard-won information. OpenAI says this helps future tasks complete faster and at a quality level previously only reachable through extensive custom instructions.

Codex also now proactively proposes work. Drawing on project context, connected plugins, and memory, the agent can suggest how to start a work day or where to pick up a previous project. OpenAI gives a concrete example: Codex can identify open comments in Google Docs that need attention, pull relevant context from Slack, Notion, and the codebase, then present a prioritized list of actions.

Staged rollout

Availability varies by feature and region. The core updates are rolling out today to Codex desktop app users signed in with ChatGPT. Personalization features — context-aware suggestions and memory — will reach Enterprise, Edu, and EU and UK users soon. Computer use starts on macOS.

For developers who have used Codex only in the terminal or editor, OpenAI is explicitly pushing them to try it across the rest of their workflow — and directing new users to download the app.

In the year since Codex launched, OpenAI says usage patterns have shifted: developers start by writing code, then increasingly use the agent to understand systems, gather context, review work, debug issues, coordinate with teammates, and keep longer-running work moving. The company ties the release to its stated mission of ensuring AGI benefits all of humanity — what it describes as "narrowing the gap between what people can imagine and what they can build."

The direction is clear. OpenAI is positioning Codex less as a coding assistant and more as a persistent software-building operator, and the company says much more is coming soon.