ChatGPT agent should be able to perform complex tasks

Samuel Buchmann

18.7.2025

Translation: machine translated

A new AI from OpenAI takes on complex tasks independently and is designed to integrate into existing workflows. Agents are an industry trend. They promise efficiency, but also threaten jobs.

OpenAI has presented the ChatGPT agent which is designed to perform complex tasks autonomously. In contrast to classic chatbots, the agent not only acts reactively, but can also carry out multi-stage workflows on instruction. It is now available for Pro, Plus and Team users. Enterprise and Education customers will follow in the coming weeks. A launch date for Switzerland and the EU is still pending.

The agent is based on an AI model developed specifically for it. It combines various tools such as a visual and text-based browser, a terminal and API access. This allows the AI to read calendar entries, carry out searches on the web, book things, create files and integrate external services such as Gmail or GitHub via so-called «connectors».

For complex tasks, the AI can take anywhere from 15 minutes to an hour. OpenAI emphasises that the focus is on successful processing, not on real-time interaction. However, the long computing time raises questions about the energy efficiency of such agents.

Permission before critical actions

The agent uses a virtual computer in an isolated environment on OpenAI servers. This means the AI has no access to local data. This is intended to provide additional security. Users can interrupt, adapt or take over tasks at any time. The AI obtains explicit authorisation before performing critical actions such as sending emails or making bookings. Financial transactions are currently «» still excluded. There are further security barriers for requests that could be used to manufacture weapons.

According to OpenAI, data transmission is encrypted and the use of personal data for training can be deactivated. Nevertheless, the risk of data leaks and misuse remains - especially when integrating external accounts and processing sensitive information. OpenAI says it has therefore implemented protection mechanisms against prompt injection attacks and other misuse scenarios.

AI agents threaten jobs

The development of agents is in vogue. Anthropic presented a tool called «Computer Use» as early as last October. Microsoft, Google and Apple are also pursuing the approach of automating workflows across different tools with Copilot, Gemini and Siri.

Some companies are already using AI agents in areas such as customer service, Marketing and support. In many cases, efficiency gains are leading to job losses, for example at large tech companies such as Meta, Amazon and Microsoft. The payment service provider Klarna stated in February 2024 that the AI answered two thirds of all customer enquiries after one month - the equivalent of the work of around 700 human employees. One year later, however, the company partially backtracked: The bots were cheaper, but also worse.

Samuel Buchmann

Senior Editor

Samuel.Buchmann@digitecgalaxus.ch

My fingerprint often changes so drastically that my MacBook doesn't recognise it anymore. The reason? If I'm not clinging to a monitor or camera, I'm probably clinging to a rockface by the tips of my fingers.

Follow

---

Software

Follow topics and stay updated on your areas of interest

Follow topic

These articles might also interest you