What is an AI Agent?
Definition: It's software with a goal
Welcome to definition #2 of the AI Dictionary Detail(ed) series. This primer defines AI agents.
Do you have a thermostat at home? Or perhaps you’ve seen the 2015 movie Inside Out? At first glance, these two experiences might seem completely unrelated. But when it comes to understanding an agent, the thermostat offers a simple, real world example while Inside Out provides a colorful analogy for complex AI agents. If you can’t claim these experiences, no worries. I’ll provide the backstory to help you follow along.
The thermostat agent: smart and smarter
The basic agent is smart
To explain the concept of an agent, let’s first understand a basic agent in a home thermostat. If you have a thermostat, you likely set it to a desired temperature. Behind the face, there is a sensor that perceives the environment—specifically, the temperature in the room.
But how does the thermostat decide when to turn the heat on or off? This is where the agent comes into play. The agent, which is really computer code, interprets the information from the sensor (sometimes called a "percept" if you want to be fancy about it). Based on this perception, the agent autonomously decides to take an action—like turning on the heat—to maintain the temperature you’ve set.
In a basic thermostat, the agent’s computer code is rule-based, meaning it follows simple instructions. It only turns the heating or cooling system on or off as needed. It’s very straightforward with the agent doing exactly what it’s programmed to do, no more, no less.
The image below illustrates how a basic agent operates within a home thermostat.
From this example, there are three key features of a basic agent that we should understand before moving on to understanding an AI agent. All agents:
Perceive their environment: they gather information from their surroundings.
Act autonomously: they make decisions without external feedback.
Take action: they perform an action based on the decisions they make.
The AI agent is smarter
The AI agent in a smart thermostat, on the other hand, contains far more sophisticated computer code than a basic agent.
Let’s compare a mock coding sample for a basic agent and an AI agent to highlight the complexity of the software of an AI-based agent.
You don’t need to understand coding to appreciate the sophistication of AI agents. An AI agent is a smarter version of a basic agent. Consequently, it enables many more actions or features than simply turning the heating or cooling on or off. It has motion sensors to detect if people are home, environmental sensors to monitor humidity, light, and temperature, and Wi-Fi to communicate with other devices.
The image below highlights how AI agents expand the features in a smart thermostat.
As you can see, an AI agent offers much more than a basic agent thanks to its more advanced software. Users can go beyond simply setting the temperature to setting features like auto-scheduling and energy-saving options.
When it comes to AI agents, their list of abilities expands. AI agents:
Access tools: they can use external resources to accomplish tasks.
Plan: they can create strategies or sequences of actions to achieve specific goals.
Store information in short and long-term memory: they remember information to improve performance.
Make decisions: they can choose the best course of action based on available data.
Learn and adapt: they can improve their performance by learning from experiences and new information.
The movie Inside Out explains AI agents
To make sense of these advanced AI agent features, let's turn to the movie Inside Out for a fun and insightful comparison.
Inside Out summary
In the film, five animated characters—Joy, Sadness, Anger, Fear, and Disgust—represent the emotions inside the mind of the protagonist, Riley. These characters operate from a central place called Headquarters, where a control panel links them to Riley’s physical world, guiding her responses and shaping her development.
Inside Out skillfully alternates between Riley's internal emotions and her external experiences, offering a simple way to understand how AI agents function in the technological world.
Emotion characters as AI agents
In fact, these “emotion characters” can be thought of as agents. First, they perform the three core features of agents. They:
perceive the environment (the external world where Riley lives),
act autonomously according to their goals (each emotion influences Riley’s emotional state), and
take actions in response to the physical world (Joy keeps Riley happy, Fear keeps her safe).
Second, the characters also illustrate the advanced AI agent capabilities that we saw in the smart thermostat. They access tools, plan, retrieve memories, make decisions, and continually learn and adapt much like AI agents do in the world of code.
Comparing AI agents to their movie counterparts
Let’s compare the list of expanded features of AI agents to their character counterparts in Inside Out.
Access to tools
When the emotion characters need to manage Riley’s feelings or responses to situations, they sometimes need additional tools to add to the console table to affect their desired outcome. These can be key or lightbulb shaped input tools from their resources in headquarters. Using these extra tools help them achieve their goals.
The AI agent also has access to tools. AI agents can access external resources like databases (e.g., weather services, utility company data) and other devices (e.g., a smartphone, security system, voice assistant). In a smart thermostat, for instance, the AI agent can access weather databases to better adjust the settings.
Planning capabilities
When it’s Riley’s first day at her new school, Joy illustrates planning capabilities when she asks Fear for a list of the potential negative outcomes; instructs Disgust to make sure Riley stands out but also blends in; and requests the “Train of Thought” to arrive with extra daydreams in case Riley gets bored in class.
AI agents also devise plans by using specialized planning algorithms in their software. This allows AI agents to improve the outcomes of their specific tasks. The "planning" involves anticipating future needs and adjusting actions based on changing conditions or new information. For instance, a smart thermostat might plan to adjust temperatures based on electricity rates or raise the temperature if the weather forecast predicts a cold front.
Short and long term memory
In Inside Out, translucent orbs represent Riley’s memories. Throughout the day, as memories form, they accumulate and reside in the main headquarters area, like a computer’s short-term memory or temporary storage. At the end of the day, Joy sends these memories off to long-term memory, where they are stored more permanently. These memories can be retrieved in the moment or later to manage Riley’s reactions or emotional development.
AI agents operate similarly, using both short-term and long-term memory to store and retrieve information. In a smart thermostat, for example, short-term memory might include the current weather conditions and the immediate home environment to help the thermostat make real-time adjustments. Long-term memory could store historical data, such as patterns of energy usage or past temperature preferences. The thermostat can retrieve that information to make more informed decisions. These memory functions are part of the agent's code.
Make decisions
The emotion characters continually make decisions that influence Riley's behavior and emotional state. For instance, Disgust decides Riley’s outfit for the first day of school, so she will fit in. Fear steps in to keep Riley protected from physical harm. The characters make these decisions based on how they perceive Riley's experiences. Their goal is to manage her emotional well-being.
AI agents also make decisions. They use the information they perceive from their environment to match their objectives. The nature of these decisions depends on the type of coding instructions. Some code is very specific. The decision will follow the rules, such as turning on the heat when the temperature drops below a certain point.
In contrast, other decisions can be more complex and involve algorithmic code that analyzes patterns. For instance, a smart thermostat might decide to adjust the temperature gradually throughout the day based on the daily schedule it has learned. AI agents could decide to pre-heat the house before you wake up or cool it down just before you arrive home from work.
Learns and adapts
A pivotal moment in the movie is when Joy learns to appreciate the importance of Sadness in Riley’s emotional journey. Initially, Joy tries to keep Sadness completely out of Riley’s experiences. She even confines Sadness to a small circle in the library, telling her never to move outside it. Over time, Joy realizes that sadness is a necessary emotion. She learns to embrace sadness as part of Riley’s emotional life and adapts to working together with Sadness.
AI agents can also learn and adapt their behavior based on previous experiences. They analyze data to identify patterns and refine their decision-making processes over time. For example, a smart thermostat learns a household's temperature preferences and adjusts its settings to maintain comfort while also being energy efficient.
The emotion characters in Inside Out help explain how the advanced software of AI agents enables enhanced features. The parallels between the movie and a smart thermostat highlight how both worlds—one fictional and one technological—operate to manage and achieve their respective goals.
AI Agents are getting even smarter
Agents Beyond the Thermostat
We have analyzed the AI agent using thermostat technology to walk us through what an agent is. But to note it, you can find agents in many technologies such as self-driving cars, personal assistants, and robotic vacuums. They behave and have characteristics like the smart thermostat.
Agents in Digital Applications
Agents, though, also exist in “digital applications” that aren’t based in a physical technology. In these applications, the agents perceive their environment by interpreting user queries. For instance, you ask a digital assistant to add a birthday dinner event to your calendar. The software code understands your request and performs that task for you.
That’s helpful, but agents are getting smarter and smarter so they can do even more complex tasks for you. Imagine your digital assistant not only adding the birthday dinner to your calendar but also selecting a restaurant, making the reservation, and even checking if your friends are available on the date and time you prefer.
Digital agents working for you
Even though the agent is digital based, and not in a technology, it’s still doing all the agent-based features we saw in the smart thermostat. For example, the agent first perceives its environment by understanding your voice command to schedule a dinner. Then, it acts autonomously, taking care of the details without needing further instructions from you.
The agent takes action by searching for restaurants that match your preferences and accesses tools like a restaurant booking service to find available reservations. It plans the best time and place for the dinner by considering your schedule and your friends’ availability.
The agent stores information in short and long-term memory and remembers your past preferences to suggest options. It makes decisions by choosing the best restaurant and time that fits everyone’s schedule. Finally, the agent learns and adapts over time, improving its recommendations the more you use it.
Why AI Agents Matter
Multi-agents: the next big step
Understanding AI agents is important as they become more sophisticated and integral to our daily lives. As agents evolve, they can do more than just automate simple tasks. In fact, just as the emotion characters in Inside Out work side by side, several AI agents can also work together.
The current trend in AI development is the rise of multi-agent systems. These systems let several specialized agents work together and accomplish various tasks at the same time. An e-commerce service like Amazon, for example, uses a multi-agent platform. It includes customer service agents, fulfillment agents, and personalized recommendation agents all working behind the scenes.
Agents in your everyday life
So why should this matter to you? AI agents manage tasks across many applications, often without us even realizing it. The thermostat in your home, once a simple device, has evolved into a smart agent that learns from your preferences, adapts to your schedule, and improves efficiency.
As AI continues to advance, these agents will only become more integrated into the technologies and services we rely on—both at home and at work. Now that you understand AI agents, keep an eye out for them—you might be surprised at how often you come across them in your daily life.