{"id":47249,"date":"2025-07-21T12:00:00","date_gmt":"2025-07-21T03:00:00","guid":{"rendered":"https:\/\/designcompass.org\/?p=47249"},"modified":"2025-07-21T08:38:10","modified_gmt":"2025-07-20T23:38:10","slug":"chatgpt-agent","status":"publish","type":"post","link":"https:\/\/designcompass.org\/en\/2025\/07\/21\/chatgpt-agent\/","title":{"rendered":"ChatGPT Agent that uses the web like a human is released"},"content":{"rendered":"

ChatGPT integrates agentic capabilities, allowing it to interact with websites and perform complex tasks as if it were a human being, using its own virtual computer. Users simply give instructions, and ChatGPT finds the information it needs, executes it, and refines the results.<\/p>\n\n\n\n

The core of this feature is the integration of two technologies that were previously provided separately. Specifically, Operator, which interacts with websites by clicking and typing, and Deep Research, which provides in-depth information through multi-level inference, are combined into a single agent. Now, ChatGPT uses the strengths of both to actively explore websites and summarize necessary information or output them in the form of slideshows and spreadsheets that organize the results.<\/p>\n\n\n\n

Even in corporate work, ChatGPT agents can perform specialized tasks for investment banking analysts, such as financial modeling for Fortune 500 companies or developing corporate acquisition models, and have shown performance far ahead of deep research or existing GPT-4o in terms of accuracy and official usage criteria.<\/p>\n\n\n\n

ChatGPT agent allows users to intervene in the task at any time during use, changing directions or adding detailed instructions when necessary. During this process, previous work is not lost and remains intact, allowing flexible collaboration. Also, it never enters sensitive information or performs actions such as reservations or purchases without the user's explicit consent, and always requests confirmation.<\/p>\n\n\n\n

It has safety nets in place, including prompt injection protection to distinguish malicious instructions, anti-abuse policies, and browser data deletion to protect privacy. In \u2018direct control mode\u2019, where users directly control the browser, ChatGPT is designed to be unable to see what they type.<\/p>\n\n\n\n

Currently, ChatGPT agent is available to Pro, Plus, and Team users, and will be available to Enterprise and Education plans within July. Pro users can use it almost unlimitedly, while general paid plan users can handle up to 50 tasks per month.<\/p>\n\n\n\n

\"\"
(C)OpenAI<\/figcaption><\/figure>\n\n\n\n
\"\"
(C)OpenAI<\/figcaption><\/figure>\n\n\n\n
\"\"
(C)OpenAI<\/figcaption><\/figure>\n\n\n\n
\"\"
(C)OpenAI<\/figcaption><\/figure>\n\n\n\n
\"\"
(C)OpenAI<\/figcaption><\/figure>\n\n\n\n
\"\"
(C)OpenAI<\/figcaption><\/figure>\n\n\n\n
\"\"
(C)OpenAI<\/figcaption><\/figure>\n\n\n\n
\"\"
(C)OpenAI<\/figcaption><\/figure>\n\n\n\n
<\/div>\n\n\n\n

More and Sources<\/h2>\n\n\n\n