What is an AI Agent?

When you use an AI chatbot such as for instance ChatGPT it tells you what you need to do in order to achieve something. This is because ChatGPT fundamentally is a brain with ears and a mouht, but no arms or legs to use an analogy.

An AI agent, is the opposite; It's about you telling the AI what you want it to do. The difference might seem small, but if you think of ChatGPT like a brain with two ears and one mouth, an AI agent is the equivalent of giving that same AI a body, such that it can do things in the real world.

AI agents are capable of performing tasks, like a personal assistant

What can an AI agent do?

Technically, robots from science fiction movies are AI agents, so any upper boundary doesn't really exist. However, to get your creative juices flowing, I'll give you a list here of what types of AI agents we have created in AINIRO previously.

• Scrape websites and create personalised marketing emails
• Retrieve articles and create social media posts
• Manage our Kubernetes cluster
• Create scheduled tasks that executes in the future, performing some sort of "AI action"

At least in theory, an AI agent can do anything you can do with a computer, as long as it has access to an API. An AI agent is basically giving control over one or more computers to the AI such that it can autonomously perform tasks on behalf of you.

The API

API implies "Application Programming Interface" and is what allows one program to interact with another. All modern software systems have APIs. This allows you to assemble AI agents that are using for instance HubSpot's API or Google APIs to perform some action. However, if you want to have a completely custom action, you will need to create your own API. In addition, a lot of existing APIs are not tied together such that they're easily accessible for your AI agents. In case you didn't notice ...

AI agents are actually just a new word for AI workflows

You see, and AI agent is actually just an AI, typically an LLM, that consumes an API, and we've already used these for a long time - The only difference being that we call these AI workflows here at AINIRO. Below is an example of what you can achieve with an AI agent that is consuming an API.

If you need an API to create your own personal assistant, and/or AI agent, and/or AI workflow - The easiest way to get started is by creating an API using Magic Cloud. The point being that with Magic you can create an API in seconds, often without any coding skills at all, allowing you to rapidly create AI agents that somehow interacts with the world.

The future of AI Agents, the Smart Home

Imagine you're on your way home from work. You call up your AI on the phone, and ask it to start making dinner, and have a freshly made Cappuccino ready in 15 minutes. There's no robot in your home, in fact, your home has an API, you can indirectly access through a speech recognition capable LLM model such as ChatGPT, translating your words into API invocations, that invokes your home, and remotely starts the process of making dinner and creating coffee.

The above might seem like science fiction, but it's actually quite easy. Give me a kitchen and a coffee machine with an API, and I'll create it in 5 minutes for you, live on YouTube, filming myself assembling the entire thing together.

The above is just an example of course, and smart coffee machines with an API are probably too expensive for most to afford - But the above example illustrates the capabilities of AI agents.