What are AI agents?

It seems that not a day goes without a story about a massive breakthrough in the tech industry. Recently, many of those stories have been focused on a new(ish) development: AI agents. They’re the newest addition to the ever-growing tech stack and some experts really think that this technology will be the one to change the world. You can forget about the last technology that was supposed to change the world, now it’s this one. 

But what are AI agents, what can they do and what are the odds they’ll take your job away? 

What is an AI agent? 

The term “AI agent” is a new one, but it describes a concept that’s been around for years. An artificial intelligence agent is a form of technology that can autonomously complete tasks for a user. Trained on reams of data through machine learning and large language models, they can respond to commands, make decisions and achieve specific goals when prompted. Some are so smart that they don’t even require a prompt to complete their assigned missions. They are, in some ways, the world’s most illustrious butler.  

What are the five types of AI agents?

Depending on where you look, you might see that there are five or seven (or more) types of AI agents. In reality, most agents fit into several of these categories, so it’s important to take these delineations with a grain of salt. For simplicity’s sake, we’ll stick to the “big five.” 

• Simple Reflex Agents: These agents are conditioned to react to things in their environment but aren’t complex enough to learn or adapt to such changes. They act and then react, but not much else. 
• Model-Based Reflex Agents: These are a step above simple reflex agents. They can respond to changes or events in their environment and can also track these changes, store them as data and learn from them to handle more complex decision-making. 
• Goal-Based Agents: As their name suggests, these agents are developed to achieve goals. They’ll take specific steps to reach them and can evaluate the outcome of these steps to understand which will enable them to achieve a given goal, meaning they’re more well-equipped to prioritize and adapt than reflex agents are. 
• Utility-Based Agents: These agents don’t just work toward goals—they also decide which goal is the most useful by using a system that weighs pros and cons. Smart thermostats are utility-based agents: they can decide whether or not to prioritize saving energy or keeping a home comfortable based on a user’s preference, the time of day, or some other list of criteria. This makes them great for tasks like managing resources or balancing different needs.
• Learning Agents: The most advanced of all the AI agents, these are capable of optimizing themselves through experience. They can learn from past data and adjust to new environments, enabling them to solve multi-step problems, setting them apart from their counterparts and making them versatile tools for complex tasks.

This kind of taxonomy isn’t rigid, though. There are agents out there that fit into more than one of the above categories. Take self-driving cars, for example. They’re goal-based in the sense that their purpose is to transport riders from one location to another safely, but they’re also utility-based in the sense that they aim to complete their task in the most efficient way, learning from past rides to understand how to avoid traffic or tolls. 

What does an AI agent do? 

As autonomous systems, the possibilities for what AI agents can do are practically limitless, their capabilities wide-ranging. You already know they can drive cars and operate robotic vacuums. AI agents are also the “computer” you might be playing against in an online chess game; they’re Siri and Alexa in your phone or in your home; they’re robots traipsing around factory floors; they’re as simple as automatic doors or as complex as surgical robots. 

To illustrate the difference between today’s AI capabilities and the supposed ones of AI agents, consider making a restaurant reservation. Five years ago, it went like this: search for restaurants offering a certain cuisine at a certain price point in a certain neighborhood, identify one, request a reservation by calling or through an online booking site or app, then repeat the last two steps if the chosen restaurant doesn’t have availability. Now, you can shorten the process by asking an AI chatbot to identify all the restaurants that fit within those same parameters. You still have to make the reservation yourself by calling them or using an app, but the process is quicker. With an AI agent, though, all you’d have to do is provide a single initial instruction and it’ll do everything for you, from researching restaurants to identifying a match to booking the reservation.

There’s a lot involved there: sifting through restaurant locations and menus, identifying a potential match and attempting to book a time, encountering the possibility of a lack of availability and understanding this blocker and then returning to the task again to find a match. This level of autonomy, agency and automation is what’s to come with AI agents, probably.

Who is building AI agents?

In short, the AI agents of today are mostly being built by the large AI companies we hear so much about. Microsoft’s Copilot already has some agent capabilities and, just recently, OpenAI launched its first agent, Operator. Some non-AI companies have also made their own, including Workday and Salesforce. But there’s a growing trend in businesses not simply providing these tools to users but enabling them to build their own. 

For example, Dassault Systèmes’ partner Mistral AI has a platform where users can create their own AI agents. Google’s Vertex also offers this function. In the same way that ChatGPT has a feature that allows users to create their own personal iterations of the chatbot easily, Mistral enables this with agents. We’re heading toward a world in which you might have a custom agent for your job, plus one (or two or 10) for handling personal matters outside the office. Technology companies are taking concrete steps to put these tools not just in the hands of developers but the average Joe and Jane, too. Turns out, we’re all technologists. 

What do AI agents mean for the workforce? 

It’s true that AI agents could remove some of the need for workers to complete automated tasks. AI systems can handle more than just rudimentary functions. They can perform predictive maintenance, lowering the risk of machine failure and therefore reducing operational costs. They can write entire blogs (like this one, except not actually this one) with the click of a button. They can be the backbone of entire businesses and industries, from manufacturing to healthcare to marketing and more. 

At the same time, though, we’ve all noticed that AI is far from perfect. Artificial intelligence can do a lot, but not on its own. 

“AI technologies should be assistive, not autonomous,” explained Marketing AI Institute CEO Paul Roezter in a recent webinar. “We believe humans remain accountable for all decisions and actions, and this becomes very important as AI agents become a more viable technology,” he added.

Of course, AI agents will have some impact on the workforce. In some industries, an agent’s capabilities could really be revolutionary, optimizing and achieving like never before. Let’s consider briefly what they could mean for two major industries: manufacturing and healthcare. 

AI in manufacturing 

With AI agents, supply chain management, for example, could become even more agile. 

“A large, complex supply chain may contain, quite literally, billions of possible combinations for optimizing decisions, and traditional methods can’t maintain pace efficiently,” explained Brian Carboni, global marketing director for consumer industries at Dassault Systèmes. The number of variables and moving parts that power the factories of today make them highly vulnerable to human error, which also makes them a significant opportunity for AI. 

Carboni’s vision for supply chains and factories of the future is tech-driven. “They’ll have advanced robotics, 6G connectivity and AI-driven maintenance, all enhanced by virtual twin technology,” he said. His idea might just be spot on, given how the manufacturing industry has already begun to embrace AI. In a case like this, what we’ll likely see is a shift in worker skill sets: laborers who once operated machinery will now need to be upskilled and trained in how to manage the technology that powers that machinery. It doesn’t mean laying off entire teams – large swathes of people who understand the inner workings and minute details of a company – but empowering those teams to adapt to new norms. 

What’s important to remember in scenarios like these is that even with the infusion of technology into the workplace, human workers are still essential. Empowering an AI agent to make decisions on the fly according to historical data could have serious potential for businesses. But it doesn’t mean replacing actual workers. The human in the loop is, has been and will be an incredibly important component of any AI-enabled process. 

AI in healthcare 

Despite some resistance to the integration of technology into healthcare, the need for artificial intelligence in the field is becoming ever clearer at every level. The National Institutes of Health found as far back as 2016 that physicians were overly bogged down with paperwork, reducing how much time they could provide patient care. Those findings were confirmed through a 2024 survey by the American Association of Family Physicians, which found upward of 80% of doctors found the time required to complete paperwork impeded their ability to provide medical care. 

Clearly, there’s a use case for artificial intelligence in the field. Providing a solution that enables automation in paperwork would allow physicians to spend more time doing the job they’re trained to do: providing medical care. And on the patient side, there’s one, too. Consider the following scenario: a patient completes a follow-up questionnaire after a check-up or procedure, noting an abnormal symptom. Due to a busy schedule, the physician reviews it later, risking delays that could worsen the condition. With an AI-enabled form, an abnormal answer to a question would trigger an immediate alert to the physician, enabling swift intervention. For patients, such a system would create a standard of care far beyond what’s available now; for physicians, it would enable medical provisioning that could save lives and effectively transform healthcare. It also wouldn’t replace any jobs but instead, make existing ones easier to do. 

On a larger scale, AI agents in healthcare could also enable swifter clinical trials, leading to a faster track to drug development and deployment.

“Building a clinical trial has so many moving parts,” explained Jeremy Mann, a data scientist working on AI agents at MEDIDATA. There are mountains of paperwork and protocols to determine every aspect of the trial, and an AI agent could provide both a time-saver and a respite from what would otherwise be manual work. 

“In cases like these, an agent would need to ask questions about the protocols,” Mann explained, noting that for each, the questions, fields and forms need to be case-specific. “You’d want to give the agent the ability to search through historical forms and fields and evaluate them and understand how appropriate they are and how suitable they are for use, then edit them accordingly.” 

An agent wouldn’t just need to know how to create a form or protocol, they’d need to be able to learn from past examples, understand the constraints and conditions of the new trial, take appropriate action to address those differences and even ask clarifying questions. Empowering a tool to do this could significantly reduce the time expenditure needed to start the process of building a trial. The quicker that process is, the faster patients in need can access life-saving drugs and therapies. 

A human-led, AI-powered future 

AI agents represent both a significant breakthrough and a challenge in the modern workforce. With their ability to automate tasks, adapt to complex demands and deliver efficiency across industries, they have already begun reshaping how work is done. From optimizing supply chains in manufacturing to streamlining administrative burdens in healthcare, these agents are driving unprecedented innovation and productivity.

Yet, their rise does not signal the replacement of human workers but rather a transformation of roles. Recent findings from Gartner suggest that even in 2028, only 15% of work decisions will be made by agentic AI, putting to bed some of the concerns about this technology grossly upending the labor market. Workers will need to upskill, adapting to new technologies while using their unique human capabilities—critical thinking, creativity, emotional intelligence and ethical judgment. Human oversight will remain crucial, ensuring accountability and guiding AI systems as they evolve.

The future isn’t AI versus humans; it’s AI and humans working together. By fostering a balanced approach that prioritizes collaboration, we can fully harness the benefits of AI agents, empowering industries while preserving the indispensable value of the human touch.