How can I help you? The rise of AI agents

The expected launch of OpenAI's agent "Operator" in January 2025 is already generating considerable buzz. Their new AI assistant is expected to perform multi-step actions autonomously, which means it can write code, book travel, and manage schedules, by interacting with other apps and services.

It’s a rare example of an AI concept generating more excitement than angst. Maybe because it’s code and not an actual robot and therefore won’t rip your arms off on a whim. You can’t watch it doing this very human thing, but you can relate to the usefulness of it doing very human tasks. Plus: agents sound cool. Maybe it’s the trench coats.

It seems a bit unreal

In October 2024, an AI agent called "Terminal of Truths" (ToT) – a name straight from the cult leader playbook - made headlines by becoming a millionaire through cryptocurrency trading. This agent, created by researcher Andy Ayrey, was given $50,000 worth of Bitcoin by venture capitalist Marc Andreessen.*

The AI then endorsed a memecoin called GOAT (stay with me; this is real), which saw its market cap surge to over $225 million. While the initial stories in the press suggested that the AI had autonomously created and promoted GOAT, the truth was inevitably less sensational: Ayrey admitted that neither he nor ToT actually created GOAT, and that the AI operates semi-autonomously with human oversight.

Nevertheless, this incident highlighted the potential for AI agents to influence financial markets and capture public imagination. And maybe the fairer comparison is not whether it did that all by itself but whether it did it better than the person you were thinking about hiring to manage your wealth.

What is it?

For the uninitiated, an AI agent is a bit like a waiter. The waiter takes your order (input), understands what you want and relays it to the chef (processes the information), and delivers your meal (output). It saves you registering on the at-table app or negotiating with the chef directly.

If you’re in a branch of Frankie and Benny’s, they’ll be immediately cheerful, patient with the kids, and handy with the allergy information. But they probably won’t be able to tell you which farm supplied the bacon. If you’re eating somewhere at the other end of the culinary spectrum – Fallow, let’s say – they will tell you the pig’s name before they changed it to ‘bacon’, and come to your table fully loaded with wine pairing choices you hadn’t even considered.

How do agents work?

AI agents are specialised software programs designed to act as personalised digital assistants. They can perceive their environment, make informed decisions, and take actions autonomously or semi-autonomously.

They do four things that a large language model, for example, can’t do on its own:

Perception: AI agents gather data from their surroundings.

Decision-making: They process this information using algorithms, making informed decisions.

Action: The agents execute actions based on their decisions.

Learning: They continuously learn from experiences.

When I started writing this blog, pi.ai suggested Alice in Wonderland was the ultimate analogy for AI agents. Unconvinced, I went back to the books. "I knew who I was this morning, but I've changed a few times since then," says Alice, as she picks through a new world, learning new rules and making choices.

Pi has a point.

Evolution at the speed of revolution

Agents are why you might have noticed chatbots suddenly improving this year – though not all of them, sadly, as my long chat history with BA customer services can attest. Classic chatbots were assembling content from banks of FAQs based on keywords and not a lot else. Which is why they didn’t quite answer your question. That’s quite different from just having all the facts in reams of data and then letting an AI promptly answer the actual question asked. Quickly, patiently, and in the house style.

This is why I love the legendary DPD chatbot story: DPD were quick out of the gate and their tech supplier hadn’t quite nailed the agent training, but you have to admit they were ahead of a lot of companies in trying to find something better than useless FAQ-based chats. And the user was getting more accurate answers than they would have before. The excess honesty and trash talk was a bonus.

What do AI agents mean for everyone?

Like actual secret agents, they’re already everywhere.

Customer service AI agents, like those used by H&M and Uber, handle enquiries and resolve issues. Customers report higher rates of satisfaction and feel more valued. But they’re popping up nearly everywhere: virtual assistants, health monitoring assistants as well as healthcare diagnostics, route management, inventory management, predictive maintenance, dynamic pricing…

In a world of chaos, AI agents bring order and simplicity. They streamline processes, improve efficiency, and personalise experiences. Siri and Alexa are AI agents that make our interactions with technology seamless and enjoyable.

The future of AI agents

It's not all sunshine and rainbows though — the rise of AI agents also brings fears about job losses. As AI agents evolve, we have to consider and manage their impact on people and industries. It’s not theoretical either: companies that are investing in AI are already laying off staff – or, more often, shelving contractors – who major in content creation or customer services. Yes, new jobs are emerging too, and there are new opportunities for companies to grow, but I’m not going to patronise a 50-year-old journalist with a family by suggesting he re-train in AI when he’s just been given his notice. I’d ask him to train the AI in the first place. After all, the best change programmes are the ones where you bring people with you.

AI agents will only become more intuitive and capable. Like Alice in Wonderland, picking her way through a strange landscape and learning the rules as she goes, they will adapt to new situations. But unlike Alice in Wonderland (spoiler alert) we won’t all be waking up at some point to find it was all a dream. Everything will change – and lots of it in ways we won’t expect.

*Funny how many self-made millionaire legends have an origin story that involved inheriting cash/an airline/property empire/emerald mine.** We might have a way to go on understanding privilege and how even AI can’t conjure money from thin air.

**Allegedly. Elon Musk disputes that last one.