What Are AI Agents?

AI agents are artificial intelligence systems that can perform a wide range of tasks and autonomously respond to changing circumstances. After receiving an initial prompt, they work on their own so that human users don’t have to walk them through the process step-by-step or constantly send new instructions.

While popular chatbots like ChatGPT can develop meal plans and provide travel itinerary ideas, AI agents take this a step further, functioning like a personal assistant: They’ll also order the groceries based on what’s in your fridge and book the flights according to your schedule. In other words, a human sets the goal, and the AI agent decides on the best course of action for achieving it.

That’s the idea, at least. Today’s AI agents don’t yet possess all of these capabilities. But tech giants like OpenAI, Google, and Microsoft, along with some well-funded startups, are working hard to make this vision a reality.

What Are AI Agents?

AI agents execute tasks on a user’s behalf without the need for constant human involvement. Some operate in the physical world through autonomous vehicles, drones and other robots, navigating their environment based on data gathered from sensors. Others are purely software-based, operating inside computer systems. Either way, an AI system is considered “agentic” if they can tackle complex tasks without explicit supervision or direction, according to researchers at Princeton University.

“The agent is a unique entity that has agency,” Sergio Gago, managing director of AI and quantum computing at financial services firm Moody’s, told Built In. “Instead of telling them what to do or giving them a question to get an answer, you give them a business goal or a very specific goal.”

Added Gago: “You allow them to do the planning.”

Related ReadingWhat Is Robotic Process Automation?

How Do AI Agents Work?

No two AI agents are alike — they all look and act differently depending on the task at hand. Generally, though, they operate through a series of interconnected components that enable them to act autonomously, without fixed rules or constant intervention. 

1. The AI agent uses sensors to collect data: After it has received its instructions from the user, an agent collects data, using sensors to perceive its environment. Sensors can range from cameras and radars for self-driving cars, to web scraping tools for virtual assistants.
2. The AI agent studies the data to understand its environment: The agent then processes this data and analyzes it, forming a model of its surroundings to determine what to do next. In some cases, the agent can also interact with other agents to access and exchange information. The data is often stored and managed by the agent in its learning system, allowing it to improve and refine its strategy going forward, as well as adapt to new scenarios. 
3. The AI agent thinks of the best solution: The agent comes up with ideas for how to tackle the task at hand, selecting the best possible solution based on its programming and learned experience. This is where the control system comes into play, leveraging algorithms like decision trees and linear regression — or, in more advanced systems, neural networks. Think of it like the agent’s “brain,” where it processes the information gathered by the sensors and brainstorms is the best course of action.
4. The actions are carried out via actuators: The chosen actions are then performed by actuators, or components that allow the agent to actually interact with its environment and take action. These can be physical mechanisms like robotic arms and wheels, or digital processes like text generation or intelligent automation. Many of the software-only AI agents on the market now leverage large language models (LLMs) to understand their goals and complete their tasks.
5. The AI agent completes the task: Once it has accomplished a job, the agent checks it off the list and moves on to the next. 
6. The AI agent works until the job is done: As it moves along the list, the agent assesses its progress by gathering information from both external sources and its internal logs. It will continue this process — collecting data, creating and completing more tasks — without pause until a human intervenes or the job is done.

More on Artificial IntelligenceExplore Built In’s AI Coverage

How Are AI Agents Used?

AI agents can be deployed across a wide range of industries and workspaces. Here are some of the most prevalent use cases:

Customer Service

One of the most common applications of AI agents is in customer service, where they are used to interact with customers. They can autonomously access internal databases, provide relevant answers to questions and even take actions like scheduling appointments or placing orders.

“AI agents can talk to customers just like humans, ask questions, collect information and resolve their problems,” Fergal Reid, vice president of AI at customer service software provider Intercom, told Built In. “They’re built to handle the common, repetitive questions customers may have, like ‘how do I reset my password?’ or ‘what is your return policy?’”

Administrative Support

AI agents can be used as a sort of personalized executive assistant. They can draft and triage emails, schedule meetings, transcribe calls and much more — all with little or no prompting from a human user.

Software Engineering

Artificial intelligence has long been a fixture in software engineering, aiding in tasks like code generation, code translation and error correction. AI agents can take things a step further by independently building and deploying entire web applications — using the same tools, tricks and steps a human software engineer would without actually having to be asked to do so by a human.

Put simply: AI agents aren’t just meant to be a coding assistant for software engineers, but software engineers themselves.

Recruiting

AI agents can streamline the hiring process by sifting through applications automatically, using machine learning to compare the data from job descriptions to the information shared on candidates’ resumes and profiles. 

When they identify people they like, the agents can schedule interviews and follow-up with them as needed. Then, after a candidate has been hired, agents can help onboard them, combining data across HR, payroll and IT to ensure they complete the proper paperwork and training.

Autonomous Vehicles

Autonomous cars and trucks equipped with AI agents can navigate and avoid obstacles in real-time, ideally getting from point A to B without any human intervention at all. But the jury is (literally) out on how safe these vehicles actually are. They still require vigorous development and testing before they can be widely adopted on roads worldwide.

More AI ToolsAI Search Engines to Know

Benefits of AI Agents

Improved Productivity

Current AI products often require users to constantly define each problem and painstakingly check their work. Ideally, AI agents minimize this level of oversight, allowing users to outsource tasks entirely and redirect their focus to jobs that require more creativity and problem-solving skills. 

Reduced Costs

By delegating certain tasks to AI agents, businesses can reduce some of the costs associated with the inefficiencies and errors that come with having an entirely human staff. Plus, AI agents’ use of predictive analytics can further enhance their productivity, helping to streamline business operations even more.

Efficient Decision-Making 

Automation technology is faster and more efficient than humans. Many AI agents can gather and analyze massive amounts of data in real-time, enabling employees to work much more strategically.

“As a human, now I have an army that helps me do a lot of stuff — not necessarily faster, but much, much deeper and much, much more precise than anything I could do on my own,” Gago said. “You can go much further and do many more things.”

Drawbacks of AI Agents

Prone to Mistakes

Many AI agents are powered by AI models, which are prone to making mistakes. And when a piece of software is taking real-world actions on behalf of a user, one mistake can have big consequences.

“It’s just like a human employee,” Flo Crivello, founder and CEO of AI agent maker Lindy, told Built In. “It screws up.”

This is why humans still need to be actively involved in the process; they can’t relinquish total control or turn their attention away from it for too long.

“The human has to be in the loop in several parts of the process — on the evaluation of the planning, on the output, but also in the middle,” Gago said. “As a human, it’s really important that you validate the plan and validate the output of the system.”

Limited Capabilities

AI agents are still a long way from being the universal assistants we see in science fiction, capable of managing all aspects of one’s personal and professional life. But the field is evolving quickly — especially on the software side with the development of increasingly powerful AI models, which perform better than ever before on a much wider range of tasks.

Potential for Misuse

AI agents have the potential to be misused for nefarious purposes, such as spreading misinformation, conducting cyberattacks, and invading privacy. At the end of the day, it’s just as easy to build an AI agent that scams the elderly out of their money as it is to build one that streamlines business operations.

“The most exciting, but dangerous and concerning prospect of AI is that it empowers everyone,” Crivello said. “Some people want to do evil things. So if you give them great power, we don’t know what’s going to happen.”

To mitigate these risks and help ensure the ethical use of AI agents, it is important that the companies making them adhere to all existing legislation and deploy their technology responsibly.

Learn From the ExpertsYour Business Automated. Now What?

Types of AI Agents (With Examples)

AI agents can be organized into five different types, each with their own unique capabilities and use cases.

1. Simple Reflex Engines

Simple reflex agents only respond to the current situation they have been trained to handle based on a set of predefined rules. They make a narrow set of decisions based solely on the current input from their environment, so they don’t remember the past or anticipate the future. If a condition is true, an action is taken. 

• Example: Smart thermostats respond directly to the given temperature of a room based on predefined rules. If the temperature drops to a certain degree, it automatically turns on the heat; if it rises above a certain degree, it automatically turns on the air conditioner.

2. Model-Based Reflex Engines

Model-based reflex agents build and maintain their own perception of the world, gathering information about their environment and the ways their actions affect it. By considering both current inputs and past information, these agents can develop an understanding of how their environment evolves, which enables them to anticipate the future and adjust their actions accordingly instead of just following predefined rules.

• Example: Self-driving cars often use model-based reflex agents to navigate roads. Equipped with various sensors, the cars keep track of the obstacles and changes in their environment, such as traffic lights ahead and vehicles nearby, allowing them to perceive and respond to their surroundings in real-time. 

3. Goal-Based Agents 

Goal-based agents are driven by specific objectives, whether that be to win a game or navigate a room. While their goals may be predefined, the rules they have to follow aren’t. In addition to evaluating real-time data in their environment, these agents consider various different actions before performing them, ultimately choosing what gets them to their desired outcome fastest and most efficiently. Every action they take is intended to get them closer to said outcome.

• Example: Goal-based agents are often used in natural language processing tasks like content generation, where they automatically produce text that most-closely resonates with an intended audience.  

4. Utility-Based Agents

Utility-based agents use complex reasoning methods to compare different scenarios and their respective pros and cons, and then choose the option that provides users with the highest overall benefit based on their preferences. These agents are useful in complex situations with multiple factors to consider.

• Example: Travelers can use utility-based agents to find the best flight tickets according to factors like their schedule and budget.

5. Learning Agents

Learning agents continuously learn from their experience in order to refine their performance and adapt automatically. They even have a sort of internal critic to evaluate past results, comparing the actions they’ve taken to the effect they’ve had on their environment.

• Example: By continuously learning from user feedback and actions, recommendation systems provide better and better suggestions to users, such as what movies to watch or what products to purchase next.

Each type of AI agent can either act independently or be combined to form a hierarchical, multi-agent system — where the higher-level agents deconstruct complex tasks into smaller ones that can be delegated to lower-level agents. As the lower-level agents complete their tasks, their progress is stored and analyzed by the higher-level agents, ensuring that they collectively achieve their goal.

Will AI Agents Take Our Jobs?

Inevitably, AI agents will alter the employment market and displace some jobs. “That’s the point,” Crivello said.

“Artificial intelligence is one of the biggest technological advancements since the industrial revolution,” Reid said. “Many industries and human work will become disrupted and look completely different than how they do now.” 

But it is still unclear when that will happen or what it will look like. 

While human workers will likely be replaced by AI agents across multiple industries, new roles in AI development, maintenance and oversight will likely be created as well. And many of the tasks that AI agents perform well are the repetitive and mundane ones that employees typically don’t want to do anyway, which could leave them with more time for the more challenging work only humans can do. 

“The people who use agents become super-powered,” Gago said. “It’s not so much about replacing people, but more about superhuman-izing them.”

All of this is to say that the impact of AI agents on the workforce is likely to result in a state of coexistence rather than outright replacement — at least for the foreseeable future.

Frequently Asked Questions