Artificial intelligence has once again seen a major update. On May 7, OpenAI officially announced that the Codex for Chrome extension is now available. This move marks the beginning of AI assistants deeply integrating into users' daily web operations.

The core logic of this plugin lies in enabling Codex to directly drive the Chrome browser. After obtaining user authorization, Codex can "read" web content, manipulate internal systems, and even assist in filling out complex forms, just like a human. This deep system call means that AI is no longer limited to dialog boxes but has become a "digital employee" capable of performing tasks on logged-in websites.

image.png

Testing Performance: Simple Tasks "Fail," Complex Challenges Show Strength

In recent function evaluations, the performance of this extension showed an interesting contrast.

To the surprise of many, when handling simple tasks such as organizing browser tabs, Codex occasionally made operational errors or failed to recognize accurately. However, when the task difficulty increased, its processing capabilities surprised the testers.

In information collection tasks targeting flagship smartphones from major brands like Apple, Huawei, and Xiaomi, Codex demonstrated high integration efficiency. Moreover, it showed significant completion rates in complex, multi-step tasks such as generating creative topic ideas, calling Deepseek to assist with writing, and automating the creation of AI news daily tables, showing strong productivity attributes.

image.png

Role Position: A Trustworthy but Monitored "Semi-Automatic Assistant"

Although it shows great potential in task processing, the current version of Codex for Chrome is not yet perfect.

Current test feedback indicates that the tool still needs improvement in accurately identifying web elements. Experts suggest that users should currently view it as a "capable semi-automatic assistant." It can greatly free up human resources when handling non-sensitive daily work, but for tasks involving core privacy or high security requirements, it still requires full-time monitoring and verification by humans.