ZAP // Dall-E 2
What if you had an effective assistant for Christmas shopping as early as 2025? “AI agents” can solve the stress you give clothes.
If you ask a chatbot to do your Christmas shopping for you this year, chances are it will send you away. “I can’t make purchases or send products directly”, is what ChatGPT responds.
But the scenario may be about to change, all thanks to “AI agents”. Ideally, these agents will work as a team, and can work together in a variety of areas — including ordering and home delivery of Christmas gifts.
In fact, one from AI creator Langchain claims that 51% of companies surveyed already use AI agents in production.
In 2024, venture funds invested around 1.8 billion dollars (1.73 billion euros) in AI agent projects. The latest from Deloitte Global Forecasts argues that 25% of companies using generative AI will launch AI agent projects in 2025.
The research company Gartner which, by 2028, 15% the decisions from work daily will be taken by agents of IA.
In October, the Anthropic — the company behind the popular bot generative AI technology Claude — launched a “computer use” what allows AI to take control of the mouse and keyboard for a user to browse and perform actions on any website.
The education specialist Leon Furze created a demonstration, using the computer to automatically navigate a learning management system, open a work page (from school, for example), create text for the work and click the submit button. All done automatically from a single text message.
More recently, Google Deepmind released its own version, which allows an AI to navigate and perform actions autonomously in the Chrome browser.
These two systems are still early releases, with Project Mariner only available to a set of trusted testers. But they give an idea of the broth that is prepared.
However, there are two significant barriers to making AI agents useful, explains computer scientist Jon Whittle in an article on .
The first, and most obvious, is the trust. Would you trust your credit card details to an AI agent?
Despite two years of advances in AI since ChatGPT, hallucinations — when AI doesn’t know an answer and just makes something up — they remain a problem.
One from this year showed that even in AI programming — one of the most popular and valuable uses of AI — 52% of AI-generated answers to coding questions they had errors.
All it takes is one AI error to send Aunt Margarida’s gift to Uncle Júlio. And let’s hope it’s a harmless mistake, like a bad gift match, and not the leak of your bank account details.
