Contents
From chatbot to digital assistant
What previously worked only as an AI-supported dialogue is becoming a real tool with Agent mode. ChatGPT can now take active action on an isolated virtual computer in the background. There, the AI clicks through websites, types text, retrieves data and interacts with graphical user interfaces as if a human were working.
Technically, the agent mode combines two existing OpenAI tools: Operator, which can navigate websites, and Deep Research, a research tool for multi-stage analyses. With the agent mode, both approaches merge, allowing ChatGPT to switch between information gathering and action in real time.
The new mode shows its strengths especially when it comes to concrete tasks that otherwise cost time and nerves. Here are a few examples of how ChatGPT can be used practically in agent mode:
- Trip planning from flights to hotel bookings
Agent mode can independently handle complex travel planning. It researches flights, compares availability, analyses hotel reviews, and automatically prepares bookings. It takes into account specifics such as luggage options, hotel location, or cancellation policies. Final approval always comes from the user. - Targeted online shopping
For specific product searches, ChatGPT can scour webshops according to certain criteria. Price limits, design preferences, reviews and shipping details are taken into account. Suitable products are presented in a structured format and, if requested, added to the shopping cart, including direct links for the final purchase. - Restaurant research with reservation
The agent mode also supports the search for suitable restaurants. It analyses reviews, availability and menu options, pays attention to special requirements such as vegetarian cuisine or accessibility, and books directly via the provider's website if required.
Users retain control
Despite all automation, the human remains number 1 in the decision loop. Before ChatGPT sends an email, transfers a form, or completes a booking, it actively seeks user consent. Highly sensitive actions such as financial transactions or legal advice are fundamentally excluded.
If desired, users can take over the browser at any time, cancel tasks, or only specifically approve certain actions. OpenAI thus wants to ensure that the agent mode is not a black box, but remains transparent and controllable.
Also interesting for workflows
In addition to private everyday tasks, the Agent mode can also be integrated into professional workflows. ChatGPT can, for example, perform competitive analyses, create presentations or evaluate data in tabular form. Via various interfaces, the agent can be directly connected to other tools such as calendars, email inboxes or project management systems.
In early tests, the agent mode proved significantly more powerful than previous GPT models and, according to OpenAI, exceeds human benchmarks in many areas.
Coming to Germany soon?
The new Agent mode is currently only available to users in the UK who have a Pro, Plus or Team subscription. OpenAI has not yet announced when exactly the feature will launch in Germany. According to OpenAI, the launch is expected to follow soon.
Source: OpenAI press release





Comments