AI agents: "Reserve a table" is lame
If AI agents can do nothing but book me a table in a restaurant, I don't need them, says Eva-Maria Weiß.
(Image: Petr Kovalenkov / Shutterstock.com)
Once upon a time, it was said that plug-ins for ChatGPT would prevail. Then it was GPTs – that were supposed to be apps for ChatGPT. It's a bit of a joke, because neither plug-ins nor GPTs are used by the masses in any way.
This could be the end of the story at this point. But when it comes to artificial intelligence, the news keeps coming. And so it happens that the supposedly powerful extensions are replaced by new AI functions. Specifically, AI agents are now driving the hype. They are supposed to take over annoying tasks for us that we don't feel like doing.
What tasks are these? The presentations of new AI functions almost always focus on the same application example. Apparently, the biggest problem facing humanity is booking a vacation or even just a table in a restaurant. Whether OpenAI or Google, the capabilities of AI were repeatedly touted as making travel planning easier in the future.
Videos by heise
"Book a table" as best practice
And how well does it work? Apparently so poorly to date that there is now only talk of AI agents being able to book a table for you in a restaurant. OpenAI has just used this example in a presentation of the ChatGPT Agent. So you can tell it to find a restaurant that has a terrace and serves poached eggs on avocado bread for breakfast. And because Google Maps or TripAdvisor or ChatGPT itself and Perplexity could also do this, the agent is also able to find a free date in a calendar. And at the end, it can reserve a table on a website at my request.
The crux of the matter is that for this function to become established, we would all have to have a perfectly managed calendar. And we would have to plan our lives accordingly. But the truth is that most of us only have a few important dates in our calendars, rarely reserve tables and usually go to a restaurant around the corner that we already know on the spur of the moment. And even if the agent could take over the vacation planning: Most people travel once or twice a year.
Now you might think that such an agent will soon be able to do much more and be particularly helpful in the work environment. But this is exactly what Sam Altman himself warns against. Access to emails? Too insecure, says the OpenAI boss. Malicious actors could attack the agent and persuade it to disclose information via a simple email. The ChatGPT agent should be given as little access as possible.
So it remains the same: finding the right restaurant and possibly making an appointment and reservation will have to suffice as best practice. Does that justify the cost of such an AI agent? Not for me.
(emw)
