OpenAI is becoming more and more like Anthropic these days, with something new released almost every day.
Last night, besides launching three new voice models, they also rolled out a Chrome plugin for Codex.
The core capability of this plugin is that it can finally operate the browser you're already logged into.
Codex previously had a built-in browser, but it couldn't operate websites that required a login.
For example, your email, Xiaohongshu, Weibo, company backends, membership systems, enterprise workbenches...
This ability is somewhat similar to what many people have used with Claude for Chrome. Now, OpenAI has added it too.
It's very simple to use. First, update Codex, then go to Plugins and install the Chrome extension:
After installation, you can directly operate it by using @Chrome inside Codex. For instance, I asked it to automatically post on Weibo:
Then, a Weibo post was automatically sent, with the content entirely generated by itself:
This example is overly simplistic, mainly to demonstrate the basic functionality.
But honestly, after watching last night's demo video, the ceiling for this thing is far higher than just posting on Weibo.
The video highlighted a few incredibly powerful features: First is "Identity Sharing."
Because it runs as a plugin in your Chrome browser, it can directly access all your cookies and login states.
This means it doesn't require you to re-enter your account credentials; it can operate within your "real working environment," which is also more secure than entrusting your account to a third party.
Second is "Multi-Tab Parallelism." It can simultaneously monitor and scrape content from multiple web pages, and even execute tasks across tabs.
To showcase this ability, the video ran a few more complex use cases that really sparked my interest:
The first is "Fully Automated Sentiment Analysis."
In the demo, Codex was tasked with browsing OpenAI's official community. It didn't just glance; it navigated through pages, scraped all posts from the past week, and then analyzed which users were complaining and who was praising, finally generating a summary table right in your browser.
This kind of task that previously required writing a custom web scraper is now accomplished with a single command.
The second is "Closed-Loop Travel Expense Reimbursement."
This is extremely practical. It first navigates to the Gmail plugin to find recent travel receipts, extracting all dates and amounts.
Then, here's the kicker.
It automatically opens the expense reporting website, uploads the receipt PDFs, and fills in every single field on the form.
This type of cross-platform (email to web) automated workflow is what truly frees up your hands.
The most mind-blowing demo was the last one, "Multi-Agent Collaboration."
The video opened four tabs, running four Codex agents to play an online collaborative drawing game.
They could see what the others were drawing in real-time and then coordinate to complete a picture together.
This "group-chat" style of web manipulation leaves plenty of room for imagining future office collaboration.
This plugin processes these new tasks in a separate tab, so it doesn't completely take over your browser.
This means you can continue doing your own thing in the browser while it quietly handles those automation tasks in the background. It's not that you can't use your browser while it's running.
In short, OpenAI has now filled in the most crucial missing piece of the Codex puzzle.
In the future, what we use might not just be a chat window, but a "digital employee" that can run behind any web page.