Ingenious: ChatGPT Can Do Even More Now
Read in other languages:
Even before artificial intelligence became a buzzword, our smartphones were already multi-purpose tools. Augmented with artificial intelligence, they are even smarter, especially when they now have an agent mode like ChatGPT. I'll tell you how to use the mode, what it can do, and what it cannot do yet.
When AI experts were asked what the trend of the year 2025 would be, most of them answered: AI agents. Yes, an AI agent is not just a buzzword, but could significantly increase the benefits of AI chatbots. Of course, this applies all the more to the AI top dog, ChatGPT, whose agent mode has proliferated in many countries.
This is ChatGPT's Agent Mode
Let's clarify the most important aspects of the new function first:
What is this agent anyway?
The new ChatGPT agent from OpenAI can independently perform complex tasks on the computer without requiring you to specify every step, as with the normal ChatGPT. It "thinks and acts proactively" and selects from various tools to solve tasks independently. Specifically, it should be able to manage your calendar automatically, create editable presentations, execute code, and carry out in-depth research on websites. You tell the AI agent what you want it to perform, and it then runs off and works through the individual points step by step.
Who is agent mode available for?
Agent mode has been available in the USA for some time now, which means you have to invest at least $20 per month to use this mode. At this price, you can make 40 requests per month, while 400 requests per month will cost you ten times the amount ($200/month).
What can ChatGPT actually do in agent mode?
We still need to determine the exact answer to that question. For instance, it can perform complex online tasks, browse websites, press buttons, compare products, and even download files. However, it not only accesses websites, but also emails! You can intervene at any time to correct the course taken if necessary.
Although OpenAI has pointed out that protective measures have been implemented, please be aware that this is still an early, unfinished product. Even OpenAI wrote on its website:
ChatGPT agent is still in its early stages. It’s capable of taking on a range of complex tasks, but it can still make mistakes.
The risks were also pointed out extensively. With this version, you can finally ask ChatGPT to perform actions on the network for the first time. This is why OpenAI mentioned the potential for new risks, as your data can be used. Do bear this in mind before you begin to mess around with the AI agent.
How to use the agent mode of ChatGPT
Personally, I have not yet found the function in the browser version of ChatGPT, but I was able to test it in the Android app. Proceed as follows:
- When you launch ChatGPT, simply tap on the tool button next to the input window on the left.
- The first thing you can do there is select the new agent mode.
- If you enter a corresponding prompt, the agent mode will automatically kick into action. It then utilizes the corresponding tools available and selects the GPT model that is most suitable.
First self-experiment: short trip to London
I wanted to test this agent mode more extensively over the next few days. For today, I want to give it a first try and show you how it went. As travel planning is a popular example of these agent tools, I decided to take a trip to London.
My prompt was simple: plan a trip for me. I would like to travel to my favorite city, London, next Friday for the weekend. ChatGPT then got back to me very quickly:
- Would I prefer to travel by plane, train, or bus?
- What kind of accommodation (hotel, Airbnb, or hostel) would I like, and what budget do I have in mind?
- Are there certain activities I have planned for the London trip?
I answered accordingly: Travel doesn't matter, hotel or Airbnb, please, but not more expensive than about €100/night. I also wanted to experience street food in Camden Town, followed by pub crawling or enjoying live music in a club. After that, ChatGPT has a lot of work to do. The AI took an entire nine minutes to process my request. Meanwhile, I can keep an eye on exactly what's happening. Booking.com, Airbnb, or Trainline.com — I am constantly aware of the agent's current plan.
- Also worth reading: Today is the day you conquer your fear of AI
The result: Initial enthusiasm is followed by some disillusionment
The result was very detailed. It began with the following information:
Here's a suggested itinerary for a weekend trip from Dortmund to London next Friday, August 1, 2025, based on information from train, bus, hostel/hotel providers and official Camden Market sources. Please note that all prices are examples only (as of July 25, 2025) and may be subject to change. So book early and confirm the exact times.
A little disillusionment set in after it initially looked promising. I received a well-sorted list of prices, travel times, and much more, which I thought was great — even my starting point of Dortmund is taken into account. However, I began to see on one of the linked pages that the cheapest flight price listed was outside of my desired travel period. The cheapest price is of no use to me if the plane doesn't leave until two weeks later.
The same goes for accommodation. I ended up on Tripadvisor and saw the accommodation had been booked from Sunday to Monday — suboptimal for someone who is leaving on Sunday. However, ChatGPT could not help the next disillusionment: with my price expectations, I tend to end up in a hostel rather than a hotel.
I was also recommended lots of great street food stalls in Camden Town, as well as stores where you can get 2-3 pints at an affordable price. I had hoped for specific concert recommendations, but that wasn't the case. Instead, they suggested venues where I could expect live music. At least I was redirected to a page where I could book tickets directly.
Finally, I'm given information about my return journey and a few general travel tips. I'm almost a little annoyed that I only planned this trip for this attempt, because the way it's presented made me want to pack my suitcase straight away.
My conclusion
The first run-in informed me that agent AI is not yet the magic bullet that many hype-driven YouTubers would like to tell me. At least that's the case with ChatGPT's agent mode. Honestly? That doesn't actually mean anything. First, because OpenAI mentioned that the entire effort is still buggy and in an early phase. There's no doubt that you can guess where the journey is headed, and I'm already looking forward to it.
Second, I liked the result! It took less than ten minutes between a crazy idea for a spontaneous city trip and a really good plan with lots of links, offers, price comparisons, and recommendations. Pick up your smartphone for this task and see how far you can get with Google in those ten minutes.
As I mentioned, I will continue to try this out over the next few days and find out what is already possible today. For now, however, you already have the information on how you can use the agent mode yourself, and my initial assessment.
