Beyond chatbots: the exciting future of generative AI agents

Executive summary

As impressive as generative AI models like ChatGPT may be, they have limitations that can often make them a chore to work with. Most notably, they can only do what you tell them to do, incapable of making their own decisions or adjusting their approach as they work on the task at hand. But what if we told you that there is actually a type of AI that could do all of those things and more – an AI agent?

• AI agents are characterised by a certain degree of autonomy and the ability to learn from experience.
• AI agents could have a wide range of useful applications in many different fields, including customer service, healthcare, finance, and marketing.
• “Five years from now, we could be in a totally different world, in which I send in a request to an AI, and the AI itself is doing all of the actions,” predicts Aaron Levie, co-founder and CEO of the enterprise cloud company Box.
• The increasing autonomy of AI agents raises some important ethical questions related to privacy and security.
• “Without a human in the loop that checks every action to see if it’s not dangerous, we might end up with actions that are criminal or could harm people,” warns computer scientist Joshua Bengio.

While AI agents are still in the early stages of development, we can already imagine the transformative impact they could have throughout our daily lives. One day, we may be able to hand over any task we don’t want to do to an AI and focus on our true passions instead. However, that future might still be a little while away. So, don’t forget to set your alarm clock just yet: your job is waiting for you.

ChatGPT on steroids

In recent years, large language models (LLMs) like ChatGPT have captured the public’s imagination with their uncanny knack for instantly generating human-like text, art, music, and more. But while we might generally agree that these models have some very impressive capabilities, they also have one fundamental limitation: they only operate when prompted by a human, and stop functioning once the understood task is complete. What if we could break free from these limitations? What if we could create an artificial intelligence (AI) that not only responds to our prompts but also takes initiative, manages its own tasks, and collaborates with other AI systems and humans alike?

Now, we know what you’re thinking – this sounds a bit far-fetched, doesn’t it? How could an AI possibly act autonomously without human prompting? Well, this might come as a surprise, but what we’ve just described is not science fiction, but a groundbreaking new tool called an ‘AI agent’. Often described as the next stage in AI’s evolution, AI agents have the potential to fundamentally change the way we live and work. How, you might ask? Well,  imagine having a highly intelligent assistant at your disposal who could take over all of the tasks you don’t want to do, allowing you to focus your time and energy on more creative and strategic pursuits. Surely that’s something we can all get behind.

AI agents: the next stage in the evolution of AI

AI agents are far more than just chatbots – they can make their own decisions, learn from experience, and perform complex tasks autonomously.

Before we proceed any further, let us first explain what an AI agent actually is and how it differs from conventional chatbots. By now, you’ve probably noticed that AI chatbots have a slight tendency to generate fictitious information, requiring you to constantly fact-check their output and adjust your prompt until you reach a satisfactory outcome. It can be a rather frustrating experience, often negating the productivity boost AI is supposed to deliver.

That’s not the case with AI agents, though. What sets these innovative tools apart from conventional chatbots is their ability to tackle complex, multi-step problems without constant human oversight. Once having received its goal, an AI agent uses an LLM to interpret its task, develop a strategy, and then carry out the necessary steps on its own. The workflow typically involves breaking down the main objective into smaller, more manageable tasks.

For each of these tasks, the agent formulates specific instructions for the LLM to follow.  Throughout this process, it continuously evaluates and adjusts the priority of tasks based on the outcomes of previous steps. The cycle continues until the agent either achieves its goal or determines that it’s not possible. If it can’t do the job, the process is terminated to prevent the agent from being stuck in an endless loop.

What can AI agents do?

At this point, you’re probably wondering: “But what can AI agents actually do for me?” As it turns out, quite a lot. The most obvious application of AI agents is likely in customer service. We’ve all dealt with those frustrating automated systems that never seem to resolve our problems in a satisfactory manner. By dropping an AI agent into the mix, however, this type of automated customer service is transformed. Rather than just following a pre-determined script, they can handle a much wider range of inquiries, from simple FAQs to complex troubleshooting, and even adjust their approach based on the feedback they receive from the customer.

AI agents could also help usher in the next generation of virtual assistants that will far surpass the capabilities of today’s most advanced AI assistants. You think Siri is impressive? How about a virtual assistant that doesn’t just answer simple questions but actively anticipates what you need? If you’re planning a trip, it could put together an entire itinerary based on your preferences, book everything for you, and even anticipate potential hiccups – all without asking for your input. The only thing you need to do is provide a desired destination, and the AI agent will take it from there.

Another common application of AI agents that you may already be familiar with – even if you don’t know it – is in self-driving vehicles (SDVs). Did you ever stop to think about how an SDV knows when to turn, slow down, or stop? In most caesars, they use an AI agent to continuously scan their surroundings and make real-time decisions based on ever-changing road conditions. While we may not have realised the dream of fully autonomous driving just yet, advancements in this field are what will bring it closer to reality.

Another common application of AI agents that you may already be familiar with – even if you don’t know it – is in self-driving vehicles (SDVs). Did you ever stop to think about how an SDV knows when to turn, slow down, or stop? In most caesars, they use an AI agent to continuously scan their surroundings and make real-time decisions based on ever-changing road conditions. While we may not have realised the dream of fully autonomous driving just yet, advancements in this field are what will bring it closer to reality.

Ethical implications

The idea of a world populated by AI-powered personal assistants that can reason, act, and even reproduce independently raises a host of ethical quandaries that we are only beginning to grapple with. Think about it – if AI agents are making decisions and taking actions without us constantly guiding them, what happens if they misunderstand what we want? Or worse, what if they do something that goes against what we believe is right? What if they start influencing the goals we set for ourselves, or the way we grow as people?

Then there’s the matter of privacy. After all, an AI agent will need to know an awful lot about us if they’re actually going to be useful. This begs the question: who else gets to see that information? We’re pretty sure you don’t want some company using what you tell your AI assistant to bombard you with ads. Let’s push the boat out a little further: what if your AI assistant could be compelled to testify against you in court? An unsettling thought, for sure. It becomes trickier still when you think about ethical decisions. Should your AI assistant refuse to help you do something that might be harmful? And who gets to decide what’s considered harmful anyway? It’s not like there’s a universal set of values we all agree on.

As AI agents become more human-like, we may find ourselves becoming increasingly co-dependent, confiding in them our most private fears, and seeking their counsel on matters both trivial and profound. But how can we be sure that the advice they provide is truly in our best interests? Can we really ensure that these powerful entities are not designed to disproportionately benefit the companies that create them rather than the individuals they are meant to serve?

• Unlike conventional chatbots, AI agents have the ability to tackle complex, multi-step problems without constant human oversight.
• Once given a goal, an AI agent works by breaking down the main objective into smaller, more manageable tasks and then formulating specific instructions for each of those goals.
• The potential applications of AI agents are nearly endless, ranging from customer service and healthcare to finance and marketing.
• The idea of artificial entities that can act autonomously raises some serious ethical issues – how can we be sure that their actions are in our best interest?

“With generative AI, we have the opportunity to address some of the most pressing needs of the healthcare industry.”

Munjal Shah, co-founder and CEO of Hippocratic AI

Automate anything with AI agents

There are already a number of AI agents out there that can help us automate a wide range of tasks in many different areas.

AI agents come in a wide variety of forms, each with its own unique set of capabilities and characteristics. For example, Devin is an AI agent developed by San Francisco-based startup Cognition. Hailed as “the world’s first fully autonomous AI software engineer,” Devin can perform complex coding tasks that would typically require a team of skilled programmers. Once provided with an initial prompt, it autonomously plans how to solve a given problem, writes the necessary code, and then rigorously tests and implements the solution. The implications of Devin’s capabilities could be far-reaching, as it essentially enables anyone to develop their own software even if they have no prior coding knowledge. Yes, that means even you could do it.

If you think Devin is impressive, just wait until you hear about this next AI model. Developed as part of a collaboration between tech giant Nvidia and Hippocratic AI, the AI-powered ‘healthcare agent’ can conduct independent conversations with patients on a variety of healthcare-related topics, ranging from health risk assessments to post-discharge checkups. What’s particularly intriguing is that these interactions occur directly between the AI and the patient, without the immediate involvement of human healthcare providers. “With generative AI, we have the opportunity to address some of the most pressing needs of the healthcare industry,” explains Munjal Shah, co-founder and CEO of Hippocratic AI. “We can help mitigate widespread staffing shortages and increase access to high-quality care – all while improving outcomes for patients.”

Even the corporate world is starting to turn its attention to AI agents. London-based tech company 11x has developed an autonomous AI agent called Alice that can perform various tasks traditionally handled by human sales and marketing teams. Instead of simply following a predefined script or set of rules, Alice uses multiple AI platforms to research potential clients and reach out to them via personalised messages – all without any human involvement. Alice even contains a fact-checking component, using one platform to verify the accuracy of the information generated by another. But here’s where it gets really interesting – the company plans to implement a feature that would allow Alice to conduct initial phone conversations with potential customers. Of course, this would only happen with the explicit consent of the person on the other end of the line.

Similarly, San Francisco-based AI startup Typeface has developed an AI agent that can automate the creation of personalised marketing campaigns and content, enabling companies to spin a compelling narrative that unfolds seamlessly across multiple touchpoints. “You just tell it what your goal is, and it will literally create the entire marketing journey and campaign across many moments,” says Abhay Parasnis, CEO of Typeface. “It will generate the first email to send, then it will generate a push notification to send a few days later, then it will create another reminder email and each of those is personalised right to the audience and right to the product.”

• Devin is an AI agent capable of performing complex coding tasks, enabling anyone to develop their own software even if they have no prior coding knowledge.
• Nvidia and Hippocratic AI have developed a ‘healthcare agent’ that can conduct independent conversations with patients on various healthcare-related topics.
• Alice is an autonomous AI agent that can perform various tasks traditionally handled by human sales and marketing teams, such as researching potential clients and reaching out to them via personalised messages.
• AI startup Typeface has developed an AI agent that can automatically create entire marketing campaigns across multiple touchpoints.

“Five years from now, we could be in a totally different world, in which I send in a request to an AI, and the AI itself is doing all of the actions.”

Aaron Levie, co-founder and CEO of Box

A force for good or a threat to society?

Experts are divided on the potential of AI agents – some believe they will revolutionise our daily lives, while others are worried they could cause harm.

When asked to share their views on AI agents, many industry experts agree that they could have a profound impact on society. Most notably, they’re going to permanently change the way we work. “Five years from now, we could be in a totally different world, in which I send in a request to an AI, and the AI itself is doing all of the actions,” predicts Aaron Levie, co-founder and CEO of the enterprise cloud company Box. “It will be just purely going off on its own, roaming around, kind of like a human does.”

However, not everyone is equally optimistic about how soon we’ll actually get there. “[A lot of] what’s easy for people is still incredibly hard for computers,” says Kanjun Qiu, co-founder and CEO of AI research lab Imbue. “Say your boss needs you to schedule a meeting with a group of important clients. That involves reasoning skills that are complex for AI – it needs to get everyone’s preferences, resolve conflicts, all while maintaining the careful touch needed when working with clients.”

Naturally, some experts are concerned that the development of highly smart and autonomous AI entities could lead to unintended and undesirable consequences. “Without a human in the loop that checks every action to see if it’s not dangerous, we might end up with actions that are criminal or could harm people,” warns computer scientist Joshua Bengio. “In years from now these systems could be smarter than us, but it doesn’t mean they have the same moral compass.”

• Industry experts largely agree that AI agents will permanently change the way we work, with some going so far as to predict that AI could do all our work for us in less than five years.
• Some experts are a bit sceptical about the timeline, citing AI’s lack of reasoning skills as a major obstacle.
• It will be crucial to ensure that there is always a human in the loop to prevent the AI agent from doing something that might end up hurting people.

A glimpse into the future

Tomorrow’s AI agents will be able to assist us with just about any task imaginable. What will we do with all that free time then?

Now that you have a better idea of what AI agents could do, it’s time to take a step back and consider the broader implications for the future of society. Imagine you have a digital sidekick that knows you inside and out – your likes, dislikes, habits, and goals. This AI doesn’t just react to your commands: it anticipates your needs and is ready to assist with any task, no matter how big or small. Say, for example, that you’re in the market for a new laptop. Instead of spending hours scouring reviews and comparing specs, you could just tell your AI assistant what you’re looking for. It would dive into the research for you, analyse all the options, and come back with a personalised recommendation based on your specific needs and budget. If you give it the green light, it could even place the order for you.

AI agents could also have a profound impact on the way we manage our health. The most striking example of this is providing everybody with their own personal health coach. Something that could tell you, “hey, you might want to get that cough checked out,” or “your stress levels seem high lately, how about we try some meditation?” To be sure, it won’t replace doctors, but maybe it could help catch things early or keep us on track with staying healthy. This could also potentially democratise access to healthcare and maybe even alleviate strain on our healthcare systems.

In the workplace, AI agents could revolutionise the way we approach our jobs, taking over routine tasks and freeing us up to focus on more creative, high-level work. Have an idea for a business but don’t know where to start? An AI agent can help out by writing business plans, creating presentations, and even generating product images based on your ideas. As they get smarter and more connected, AI agents may even start working with one another to tackle bigger challenges. No doubt, we will see the emergence of ‘AI agent-as-a-service’ platforms, which will connect users with the most suitable agents for specific jobs.

But as exciting as all this undoubtedly is, it also raises some big questions about the future. In a world where AI agents can take care of most of our needs, what will people do with their time? Will education still be important when you can just ask your AI for answers? And as we spend more time interacting with increasingly human-like AI assistants, will we start to form pseudo-relationships with these digital entities? Could this lead to a sense of isolation or disconnection from our fellow humans?

But as exciting as all this undoubtedly is, it also raises some big questions about the future. In a world where AI agents can take care of most of our needs, what will people do with their time? Will education still be important when you can just ask your AI for answers? And as we spend more time interacting with increasingly human-like AI assistants, will we start to form pseudo-relationships with these digital entities? Could this lead to a sense of isolation or disconnection from our fellow humans?

• In the future, AI agents will know everything about us and they will even be able to anticipate our needs.
• While they may never fully replace doctors, AI agents could act as our own personal health coaches and help us stay healthy.
• AI agents could have a significant impact in the workplace, taking over routine tasks and enabling human workers to focus on more important work.
• As AI agents become increasingly human-like, there are concerns that we may become emotionally dependent on them and isolate ourselves from other humans.

Learnings:

Clearly, AI agents represent a significant advancement in the field of AI. With their ability to tackle complex problems and execute actions autonomously, AI agents push the boundaries of the possible – especially when compared to conventional chatbots. It’s a development that has the potential to fundamentally reshape every aspect of our lives, from the way we work to the way we manage our health.

• AI agents are now capable of taking over tasks that previously required highly skilled professionals. For example, an AI agent called Devin can code a website from scratch with minimal human input.
• In healthcare, AI agents could be used to perform a wide range of tasks typically handled by nurses, including health risk assessments, pre-op check-ins, and post-discharge checkups.
• Despite their impressive capabilities, AI agents still struggle to accomplish certain tasks that are relatively easy for humans, limiting their usefulness (for now).
• Some experts warn that AI agents acting autonomously without human oversight could end up causing harm to humans by taking actions that aren’t in our best interest.

But here’s the thing – as AI agents get more and more autonomous, it raises some tricky ethical questions. What if there’s a misunderstanding between you and your AI assistant? And who else is going to have access to all the personal info you share with them? There’s also the issue of how much influence AI should have over our personal growth and decision-making. As we move forward, we’ve got to make sure that AI agents are actually working in the best interests of people and society – not just lining the pockets of the companies that create them.