X Blocked It, but AI Got Through
X (formerly Twitter) doesn't offer a way to export your bookmarks all at once.
To sort through a pile of bookmarks, all you can do is scroll endlessly.
Ethan Mollick, a professor at the University of Pennsylvania's Wharton School, shared on X at 6:22 AM (KST) on August 17, 2026, that he handed this problem straight to an AI.
All 5,302 Bookmarks Came Out
Mollick ran GPT-5.6 Sol inside OpenAI's Codex and asked it to export his bookmarks.
The AI drove his Chrome browser directly and pulled all 5,302 bookmarks, going back to 2014, along with assorted metadata.
Mollick then asked the AI to "find the gems among the pile of bookmarks," and it picked out interesting items, such as whales' learned escape behavior and a model that compressed 4.3 million recipes into 2 megabytes, and organized them into a thread.
The post has more than 81,000 views (as of August 17, 2026).
Why Couldn't Grok Bot Do It?
In the comments, a user asked, "Couldn't Grok Bot do something similar with an X plugin that has read access?" Mollick replied that he had checked it himself.
According to his reply, Grok Bot couldn't access the full bookmark list through the API. The API's supported scope is limited, so it couldn't retrieve the whole list.
Codex's GPT-5.6 Sol, by contrast, got around the limit by operating the browser screen directly instead of using the API.
Operating the Browser Instead of Writing Code
What this case shows is clear. Even when the API is blocked, if a person can do something by looking at the screen and scrolling, an AI agent can operate the browser for you and get the same result.
It's also an example that, when you need to organize data from a service with no official export, telling an agent to "look at the screen and do it yourself" works in practice, rather than waiting for an API integration.
Still, this is a one-off account of a task on a personal account. Whether it could end up circumventing the target service's terms of use or automation policies needs separate judgment.
A Use Not in OpenAI's Official Docs
This article is a hands-on account based on Mollick's own post. Whether OpenAI has documented this kind of browser automation as an official Codex feature wasn't separately confirmed in this check.
Specific work logs, such as how long the extraction of 5,302 bookmarks took and its failure rate, also haven't been released.