AI agents in plain English, and why you should care.

💡

Note: I have oversimplified a lot of concepts which may impact the technical accuracy of some details.

🤔 What is Artificial Intelligence (AI)?

AI is the pursuit of creating technology capable of thinking, reasoning, and acting as well as—or even better than humans.

You might wonder why we’re creating it. Like any other technological breakthrough, AI is a tool designed to help us accomplish tasks more efficiently. When we needed faster transportation, we hit the limits of what horses could achieve, so we invented cars. Similarly, as we confront the limits of our own abilities, we’re inventing AI to bridge the gap and support us.

Now, back to agents.

🤖 What is an AI agent?

The word agent typically refers to a person who acts on behalf of another person or group. An AI agent, therefore, is a type of software that you send to perform tasks on your behalf. An AI agent can:

Understand the task you are asking it to perform;
Determine the end result of the task;
Plan the execution of the task;
Execute the task based on the plan.

An example use of an agent would be something like:

Prompt: "Plan a family vacation to a tropical country next week and email me the itinerary. I expect good weather."

The agent will then do what you've asked.

⚙️ How do AI agents work?

AI agents are built on top of the same technology that powers applications like ChatGPT and chatbots. These are large banks of knowledge that give it the ability to read, understand and write. They are called models. So, when an agent gets a prompt like the one above, it may do the following:

Find tropical countries that won't require visas or have a visa-on-arrival program.
If it's based on an up-to-date model, it will just ask the model for answers and move on to the next task. Else, it may look online.
[output] a list of countries with a tropical climate.
From the list of countries, find a city with the best weather next week.
The model will not have answers to this, so the agent must use another service (called a tool) to retrieve weather information and decide.
[output] a city in one of the countries.
Find places of interest and build the itinerary.
Again, the agent may ask the model to provide answers if it is up to date and use that information to build the itinerary.
[output] itinerary.
Email the itinerary to the user.
This concludes the task.

💡

Note: I picked this example just to explain the concept, so it's limited by design. In real life, agents can be a lot more sophisticated. For example, they could factor in the distance from where the user is, their budget and how many people are going on the trip.

⚖️ How is this different from ChatGPT?

The two types of applications are related, but ChatGPT (or similar) were designed to be Question-and-Answer applications. They provide conversational interactions and are not expected to perform any actions.

What sets an agent apart from something like ChatGPT are the following:

Access to tools:
Agents are equipped with access to a wide range of tools, enabling them to perform tasks like sending emails, conducting online searches, checking weather updates, or even interacting with aggregated flight booking platforms.
No human involvement (aspirational goal):
Agents are designed to act autonomously from start to finish. They will still use the Question-and-Answer approach to interact with their knowledge bases, but they have to reason independently.
Agents are specialists:
An agent is trained and given instructions to excel at something specific. The tools they are given access to are usually suited to the type of tasks they can undertake. This is why in certain cases, multiple agents may collaborate to complete a task.

🦾 Can agents replace people?

We are not quite there yet. AI agents offer significant potential but still face several challenges and ethical concerns:

Understanding nuanced instructions: AI agents struggle with ambiguous or complex human input, which can lead to errors or misinterpretation.
Bias in decision-making: Training data biases can influence AI agents, potentially reinforcing unfair or discriminatory outcomes.
Accountability issues: Determining responsibility for an agent’s actions, especially in autonomous tasks, remains a challenge.
Privacy concerns: Handling sensitive user data securely and adhering to privacy regulations like GDPR and CCPA is critical to maintaining trust.

Balancing innovation with ethical practices is key to unlocking the full potential of AI agents responsibly.

☀️ Why should you care about all this?

AI agents have the potential to either revolutionize or disrupt your business, depending on how you adapt to them. Whether they present an opportunity or a challenge, now is the time to lean in and prepare.

The possibilities for simplifying workflows and eliminating tedious tasks are immense. My hope is that this article inspires you to see the same opportunities and embrace the transformative potential of AI agents.