OpenAI’s Codex just moved into the place where most of our daily work already lives: the Chrome browser on macOS and Windows. With a new Chrome extension, Codex can now act inside your real, logged-in browser session, using the same tabs, cookies, and accounts you already have open, without hijacking your screen or taking over your browsing.

Instead of being “that coding agent in a separate app,” Codex can now quietly sit alongside you in Chrome, stepping in when something is tedious, repetitive, or just annoyingly manual. The idea is simple but powerful: if a task lives on the web and depends on you being signed into services like Gmail, Salesforce, GitHub, Notion, or internal dashboards, Codex can now do a lot of that clicking, typing, and navigating for you, directly in your own browser profile.

At a technical level, the Chrome extension bridges Codex’s existing skills with your real browsing environment. The Codex app already had an in-app browser and plugins that work with specific services or APIs, but those are mostly ideal for public pages, local dev servers, and dedicated integrations. The Chrome extension is for everything else: the messy, real-world mix of SaaS tools, intranet portals, and logged-in pages where your day actually happens, from CRMs to analytics dashboards.

OpenAI describes it as a way for Codex to decide which tool is best for each step of a job. If there’s a plugin for a service, it might use that; if it needs a logged-in site, it jumps into Chrome; if it just needs to test a local app you’re building, it can use the in-app browser. Behind the scenes, Codex is writing and running code to control Chrome, meaning it isn’t just replaying recorded clicks but actually scripting flows, handling logic, and adapting when pages change.

The upgrade is also about speed and parallelism. Codex can work across multiple tabs in the background while you keep using your browser normally. Think of workflows like: filling out similar forms in multiple internal tools, stepping through a long onboarding wizard across accounts, testing the same web app flow with different users, or scraping structured data from a series of dashboards without babysitting each step. Because Codex can spin up multiple agents and bind them to different tabs, you can effectively have a small team of AI “junior operators” working on your behalf, all inside Chrome.

Developers are an obvious first audience. With the extension, Codex can test web apps in a realistic environment, poke at UI flows, hit endpoints in context, and even use DevTools – all while the developer keeps coding or reviewing changes elsewhere. For example, a developer could ask Codex to run through a checkout flow repeatedly after changing some frontend logic, or to verify sign-up works across a few test accounts, and let it handle the clicking and validation in Chrome. Because Codex can also read and edit code via the Codex app and IDE integrations, it can now debug both the code and the live browser behavior as part of the same workflow.

But OpenAI is clearly aiming beyond developers. Most knowledge workers live inside browser tabs all day, switching between email, CRM, documentation, spreadsheets, and internal tools. With the Chrome extension, Codex can log into those same systems, run searches, cross-reference data, update records, and move information from one tool to another, turning the browser into a more automated workspace. OpenAI highlights tasks like checking dashboards, updating CRMs, running research workflows, or shepherding data through multi-step web forms as good fits for this kind of automation.

On the user-experience side, the extension is designed not to “take over” your browser, which has been a big concern with earlier generations of automation tools. Instead of stealing focus or constantly driving your mouse cursor, Codex runs in the background, opening task-specific tabs and working in them while leaving your main browsing largely unaffected. Because it can rely on code execution instead of pure screen-scraping, it can interact with pages more directly and reliably, which helps reduce the jittery, robotic-feeling control that many people associate with automation.

Getting started flows through the Codex app rather than the Chrome Web Store directly. You open the Codex app, head to the Plugins section, add the Chrome plugin, and then follow the guided setup that walks you through installing and connecting the Chrome extension. Chrome will show a serious-looking list of permissions: reading and changing data on websites, accessing browsing history across signed-in devices, interacting with tab groups, and more. Those prompts aren’t fluff; they reflect the fact that Codex is effectively gaining the same level of access to your browser that a powerful user script or dev tool might have.

That leads to the obvious concerns: privacy and security. OpenAI’s documentation emphasizes that you should treat any page content that Codex sees as untrusted context and make deliberate decisions about where you let it operate. The company says it does not maintain a complete standalone log of every browser action; instead, it stores only the parts that become part of a Codex “thread” – things like page text that is read, screenshots, tool calls, summaries, and messages. Still, by design, you’re giving an AI agent both the ability and the permission to act as you, inside your logged-in web life, so you have to be comfortable with that tradeoff, especially around sensitive accounts like banking or healthcare portals.

The rollout is also geographically scoped. The Chrome extension is available through the Codex app on macOS and Windows in most regions, but not yet in the EU and UK, where support is “coming soon.” That tracks with the tighter regulatory environment in Europe around data access and automated processing, and suggests OpenAI is still aligning this feature with regional compliance requirements. For everyone else, the extension is already live, and early commentary from developers and AI watchers is that this is a meaningful step toward “real” AI agents that live where users actually work, not just in sandboxes.