Agent ChatGPT. A completely new tool by OpenAI  – AI News – #3 July 2025

Imagine if your workday started differently. Instead of opening dozens of tabs, checking emails and calendars, and then manually compiling data for a report, you simply write one sentence to your assistant: “Prepare a summary of upcoming client meetings based on the latest industry news, and turn it into a brief presentation.” After a few minutes, you receive a ready, editable file. This isn’t a scene from a science fiction movie. This is the reality that OpenAI has just made available by introducing the ChatGPT Agent.

Until now, AI agents have mostly been discussed in the context of futuristic research and niche projects. These were powerful but complex tools, inaccessible to the average user. OpenAI has decided to change that by integrating this technology directly with ChatGPT, putting an unprecedentedly powerful tool into the hands of millions of people.

In short, the ChatGPT Agent is AI that has been given its own set of tools to operate online. Instead of being limited to generating text, it can independently browse websites, click links, filter information, and even run code. It’s like handing an intelligent assistant the keys to a car and a credit card—it can now not only tell you where the best restaurant is, but also book a table and order a taxi.

This new feature is the culmination of previous OpenAI experiments, combining the best of two worlds. On one side, there are the capabilities of the “Operator” tool, which could simulate human interactions with websites (clicking, typing, scrolling). On the other—the analytical power of the “deep research” module, specialized in synthesizing massive amounts of data.

Separately, these tools were limited. Operator couldn’t perform in-depth analyses, and deep research couldn’t log into a site to retrieve fresh data. The ChatGPT Agent merges these competencies into one cohesive system. As a result, it can smoothly transition from reasoning and analysis to concrete action, executing complex workflows from start to finish based on your instructions.

The strength of the ChatGPT Agent lies in its versatility. This isn’t a single rigid tool, but a flexible system that chooses suitable resources to achieve its goals. The agent has an entire arsenal at its disposal:

This means ChatGPT Agent can, for example, check your calendar through an API, find a free slot, and then use the visual browser to book cinema tickets for a specific time—all within a single task.

The vision of autonomous AI acting on our behalf may raise concerns. However, OpenAI has ensured the user always remains in the driver’s seat. The agent asks for permission before undertaking key actions (such as making a purchase), and the entire process is transparent. You can continuously monitor what it’s doing, stop its work at any time, take over control of the browser, or halt the task entirely. If the agent doesn’t know something or needs clarification, it will proactively ask you for guidance.

Making such a powerful tool available to the broader public brings new security challenges. The ability to act on websites and handle user data introduces new vectors for potential attacks.

A key threat is so-called prompt injection. Imagine an agent, in carrying out your instruction, encounters a website with maliciously crafted code (for example, hidden in metadata). This code could “trick” the agent into performing unintended actions, such as sending your private data to a third party.

OpenAI ensures it has implemented a range of protections, including advanced monitoring mechanisms and requiring user confirmation for important operations. Nevertheless, awareness and caution on the user’s part remain essential, especially when deciding which data and applications you grant the agent access to.

The ChatGPT Agent is undoubtedly a step into the future, but this is just the beginning of the road. Although impressive, the tool is still in its early development stages and can make mistakes. Some features, like presentation generation, remain in beta and their results can be basic.

There is also the issue of speed—complex, multi-stage tasks may take the agent several, or even a dozen, minutes. However, even in this form, it is a massive time-saver compared to the hours we would spend performing these tasks manually.

We are witnessing the birth of a completely new category of AI tools. The ChatGPT Agent shifts the paradigm from passive answering to active doing. It’s no longer just your research partner but a true collaborator that can genuinely relieve you of daily tasks—both in your personal and professional life. And this is just the first publicly available version. It’s daunting to imagine what it will be capable of in a year. For sure, we’ll be following further developments in this field, so stay up to date with Delante and subscribe to our newsletter!

Source of information about the ChatGPT Agent: https://openai.com/index/introducing-chatgpt-agent/