ChatGPT can now work with its own computer to complete tasks for you. What do the first users say about the Agent feature? Home News The new feature combines the capabilities of previous Operator and Deep Research systems ChatGPT now independently works with the web, performs analyses, and creates documents according to user instructions The feature is available for subscribers, but not yet in European Union countries Sdílejte: Adam Kurfürst Published: 20. 7. 2025 04:30 OpenAI introduced the new ChatGPT Agent feature, which elevates the capabilities of artificial intelligence to an entirely new level. While ChatGPT primarily answered questions until now, the new feature introduced on Thursday allows it to independently perform complex tasks using its own virtual computer. The Agent can seamlessly transition between reasoning and concrete actions to manage complex workflows from start to finish, all based on simple user instructions. Complex Tasks in a Few Clicks Work Under User Supervision Risks and Security Measures Availability and Limitations What Early Users Are Saying Advanced AI Becomes More Accessible Complex Tasks in a Few Clicks ChatGPT Agent represents a unified agent system that combines three key capabilities: interaction with websites (previously available as part of the Operator feature), complex research and information analysis (Deep Research), and the conversational intelligence of ChatGPT itself. With this combination, it can, for example, examine your calendar and prepare an overview of upcoming meetings based on the latest news, plan and purchase ingredients for a Japanese breakfast for four people, or analyze competition and create a presentation with the results. ChatGPT can now do work for you using its own computer.Introducing ChatGPT agent—a unified agentic system combining Operator’s action-taking remote browser, deep research’s web synthesis, and ChatGPT’s conversational strengths. pic.twitter.com/7uN2Nc6nBQ— OpenAI (@OpenAI) July 17, 2025 To accomplish these tasks, the Agent has a set of tools at its disposal: a visual browser for web interaction, a text browser for simpler searches, a terminal, and direct API access. Additionally, it can connect with third-party services such as Gmail and Github. If website login is required, the user can take control of the browser and log in securely. Work Under User Supervision One of the key features of ChatGPT Agent is interactive collaboration with the user. During task execution, you can interrupt the process at any time, clarify instructions, or change the assignment. The Agent then continues where it left off, but with new information and without losing previous progress. ChatGPT itself can also proactively ask for more details if it needs them to complete the task. Let’s show you what ChatGPT agent can do 🧵Need Monday metrics? ChatGPT agent can fetch the data, generate the spreadsheet, and schedule it to run again—automatically. pic.twitter.com/oPz7bxMkUK— OpenAI (@OpenAI) July 17, 2025 It is important that the user always has control. The Agent requires permission before performing important actions, and you can take over the browser or stop the task entirely at any time. If you have the ChatGPT app on your phone, you will receive a notification once the task is completed. Risks and Security Measures Given that this is the first instance where ChatGPT can perform actions on the web, it also introduces new security risks. Particularly concerning is the possibility of manipulation through so-called prompt injection, where malicious instructions hidden on a website could induce the agent to perform undesirable actions, such as sharing private user data. OpenAI has implemented a series of security measures, including training the agent to recognize and resist this phenomenon, monitoring for rapid attack detection, and requiring explicit user confirmation before important actions. For sensitive tasks like sending emails, active supervision is required, and the agent is trained to actively refuse high-risk tasks, such as bank transfers. The feature naturally has its limitations. Understandably, it can still make mistakes, and some capabilities, like creating presentations, are currently in beta. Results can sometimes be simple and incomplete, especially when creating a new document without an existing template. However, OpenAI promises that the feature will gradually improve and refine. Availability and Limitations ChatGPT Agent will be available to users of Pro, Plus, and Team subscriptions, with Pro users receiving 400 messages per month, while other paid users receive 40 messages with the option to purchase additional credits. For businesses and educational institutions, the feature will be available in the coming weeks. OpenAI is still working on making the service available in countries of the European Economic Area (including ours) and Switzerland. Due to high demand, only Pro tier subscribers have received access so far; Plus and Team subscribers will get access on Monday. Now rolled out to 100% of Pro users. Due to higher than expected demand, Plus and Team users will begin getting access Monday. https://t.co/StDl9ieN1X— OpenAI (@OpenAI) July 18, 2025 What Early Users Are Saying Professor Ethan Mollick, who had early access to the feature, considers it a “big step forward for AIs to do real work.” According to him, the Agent does a good job with autonomous research and compiling Excel files, even with formulas, and it reminds him more of “working with a real human intern capable of a wider range of analytical and computational tasks.” I had early access & ChatGPT agent is, I think, a big step forward for getting AIs to do real workEven at this stage, it does a good job autonomously doing research & assembling Excel files (with formulas!), PowerPoint, etc.It gives a sense of how agents are coming together pic.twitter.com/XAqOjufV59— Ethan Mollick (@emollick) July 17, 2025 On the other hand, Olivia Moore, a partner at venture capital firm a16z, in her evaluation after a day of testing, points out the slowness of the Agent compared to competing products, such as the newly launched Comet browser from Perplexity. According to her, this is likely due to the product launching a “virtual computer” for each task, which is slower than a mere API call. She also mentions that the Agent often “over-engineered” simple tasks but failed to complete more complex ones. Advanced AI Becomes More Accessible It is evident that even advanced AI services and functionalities are becoming increasingly accessible to ordinary users. While Operator, capable of autonomous work in a web environment, was only available to Pro tier subscribers, costing $200 per month (an astronomical sum for a typical user), ChatGPT Agent is now included in the Plus package, which costs only a few hundred. It is likely only a matter of time before similar features, albeit certainly limited in terms of performance, also spread to free versions of popular AI platforms. What are your thoughts on the ChatGPT Agent feature? Sources: OpenAI, Sam Altman (X), Ethan Mollick (X), Olivia Moore (X) About the author Adam Kurfürst Adam studuje na gymnáziu a technologické žurnalistice se věnuje od svých 14 let. Pakliže pomineme jeho vášeň pro chytré telefony, tablety a příslušenství, rád se… More about the author Sdílejte: AI ChatGPT OpenAI Umělá inteligence