<p>OpenAI has shipped a significant update to Codex that moves it well past its origins as a coding assistant. The headline addition is computer use on macOS — Codex gets its own cursor and keyboard, so it can click through native apps like Slack, Spotify, Google Sheets, and iMessage without interrupting whatever you’re doing. It runs in the background, in parallel with your work.</p>
<p>The browser side got a similar upgrade. A new Chrome extension (rolling out on macOS and Windows) lets Codex work across tabs in the background, run web DevTools, test web apps, and gather context from multiple pages — without taking over your browser. You stay in control of which sites it can access.</p>
<h2>What computer use actually unlocks</h2>
<p>A few things stand out about how this works in practice:</p>
<ul>
<li><strong>Any app becomes a tool.</strong> Because Codex uses the screen directly, it doesn’t need an API or plugin for every app. If you can see it, Codex can use it.</li>
<li><strong>It uses its own cursor.</strong> You can keep working while Codex handles something else. No context switching, no waiting.</li>
<li><strong>It can learn from what it does.</strong> After completing a task, you can ask Codex to reflect on the steps and turn them into a reusable skill or a scheduled automation. That combination — computer use plus automations — is where it gets genuinely useful for repetitive work.</li>
<li><strong>Speed.</strong> The latest app update (v26.429.20946) claims 20% faster computer and browser use, along with upgraded slides and sheets capabilities.</li>
</ul>
<p>Sam Altman called computer use “even more useful than I expected” and flagged it as a real update for non-coding work specifically.</p>
<h2>Beyond coding: the broader picture</h2>
<p>Codex now has 90+ integrations (Jira, GitLab, Microsoft Suite, Google Calendar, Gmail, and more), the ability to schedule its own tasks, and a research preview feature called Chronicle that watches your screen, takes periodic screenshots, and builds structured local Markdown memories. That last one is early, but it points toward where this is heading: less “coding assistant,” more general-purpose AI workspace.</p>
<p>For developers specifically, the in-app browser has proven useful for pointing Codex directly at a GitHub issue — in one test against the HTTPie Python codebase, it traced a bug, wrote a fix, and added a regression test in about three minutes. PR review works too: it reads the diff, cites relevant documentation, runs tests, and flags gaps in coverage.</p>
<p>There’s also SSH support in the works. A “Remote Control” feature visible in settings will let Codex manage remote instances over SSH — which will inevitably lead to someone asking it to fix a production bug directly.</p>
<h2>How it compares to Claude Code</h2>
<p>A few developers who’ve switched report that GPT-5.5 scores 82.7% on Terminal-Bench 2.0 versus Claude Opus 4.7’s 69.4%. The pricing gap is also notable: Codex on the $20/month plan offers more usable daily runway than Claude Max at $100/month, by most accounts.</p>
<p>The other practical difference is autonomy. Codex tends to make reasonable assumptions and keep moving rather than stopping to ask clarifying questions. Whether that’s a feature or a bug depends on your workflow, but for long-horizon tasks it reduces friction.</p>
<p>One caveat: Browser Use and Computer Use appear to be disabled in the EU for now, likely for regulatory reasons.</p>
<h2>Building a second brain with Codex</h2>
<p>One interesting use case that’s emerged separately: using Codex as the processing layer for a personal knowledge system. The setup pairs Codex with Obsidian — content clipped from YouTube, articles, and podcasts lands in a raw folder, Codex processes it hourly into interlinked wiki pages, and journaling queries get answered by pulling from the wiki, past journal entries, and contact notes. Automations push backups to a private GitHub repo. It’s a practical demonstration that Codex’s scheduling and file-handling capabilities extend well beyond software development.</p>
<h2>The competitive context</h2>
<p>Codex is entering a crowded space. Claude Code, Cursor, and the upcoming Grok Build (xAI’s desktop coding app for macOS, Windows, and Linux) are all competing for the same workflows. What distinguishes Codex right now is the combination of computer use, the Chrome extension, the automation/scheduling layer, and the polish of the desktop app itself. The UI details — app icons when you @ an application, a built-in file viewer and browser — reduce the friction of switching between tools.</p>
<p>Whether that’s enough to hold the lead is an open question. But the update is substantial, and the direction is clear: OpenAI is building Codex into something closer to an AI workspace than a coding tool.</p>


Collections

OpenAI has shipped a major update to Codex that expands it well beyond coding. Key additions include computer use on macOS (Codex gets its own cursor and keyboard to operate native apps like Slack and Google Sheets in the background), a new Chrome extension for background browser automation, and 90+ integrations with tools like Jira, GitLab, and Google Calendar. A research preview feature called Chronicle watches your screen and builds local Markdown memories. For developers, Codex can be pointed at GitHub issues to trace bugs, write fixes, and add regression tests. SSH remote control is also in the works. Compared to Claude Code, GPT-5.5 scores higher on Terminal-Bench 2.0 (82.7% vs 69.4%) and the $20/month plan offers more daily runway than Claude Max at $100/month. Computer use and browser use are currently disabled in the EU. The overall direction is clear: OpenAI is positioning Codex as a general-purpose AI workspace rather than just a coding assistant.

Codex Desktop adds Computer Use on macOS