What are agents?

1. What are agents?

Before we dive into coding our agent, let’s answer a fundamental question: what exactly is an AI agent? And how is it different from just calling the GPT model directly?

2. LLMs have limitations

A standard GPT model is like a brilliant but single-minded expert: if we ask it a question, it gives us an answer from its training knowledge. This is already very helpful, but it doesn’t do anything beyond giving text responses. If it doesn’t know something, it can’t go fetch information or take actions unless we explicitly feed it more data.

3. Agents: the bigger picture

An agent, on the other hand, is a system that leverages an AI model to interact with its environment to achieve a user-defined objective. It combines reasoning, planning, and executing actions to fulfill tasks, including tasks that need external tools to be fulfilled.

4. Agents: the bigger picture

Imagine you had an AI that, when faced with a tough question, could decide whether to search the web to find the answer or to make some step-by-step calculations: that’s what an agent does. It can plan a series of steps, use external resources, and orchestrate those steps to fulfill a task. By doing so, it can handle more complex or open‑ended tasks by breaking them down and interacting with the world as needed. In essence, an agent has a bit of “agency” – hence the name.

5. LLMs vs. Agents

So when should we use a simple LLM versus an agent? If our task is straightforward and self-contained (like “summarize this text” or “translate this paragraph”), a direct LLM call is usually enough. But if the task might require fetching new information, using multiple steps, or handling conditional decisions (like “research this topic and compile a report with sources”), an agentic approach really excels. Agents are also great when we want the AI to be extensible – for example, giving it a calculator tool for math or internet access for up‑to‑date information.

6. Agentic workflows

In this course, we’re focusing on agentic workflows – essentially, we’ll be building these more capable AI assistants. We’ll start simple, with one agent and one tool, and gradually increase complexity, up to developing a deep research agentic workflow that can perform its own research and critique it.

7. Let's practice!

Exciting, right? Let’s get into it.

Create Your Free Account

By continuing, you accept our Terms of Use, our Privacy Policy and that your data is stored in the USA.