People use one word for very different software. The chatbot that answers your questions, the camera feature that spots a face, the tool that turns a sentence into an image, and the assistant that books a table are all called AI. They work differently, and they are good at different things.
What they share is one idea. Instead of following rules a programmer wrote by hand, the system learned patterns from examples. That is the whole trick, and it is enough to write, draw, talk, and act.
Learned patterns, not written rules
Ordinary software is a list of instructions. A programmer writes every rule: if the password matches, log the person in. AI is built the other way around. You show the system millions of examples and it works out the rules itself. Nobody writes down what a cat looks like. The system sees enough cats to get the idea.
This is also why AI needs so much data. A person can learn what a stop sign looks like from one example. The system needs thousands, because nobody tells it the rule and it has to work the rule out on its own. Once it gets there, though, it can apply what it learned to every email and every photo instantly, which no person can do.
One word, many machines
A spam filter, a face unlock, a movie recommendation, a chatbot, and an agent that browses for you are all AI, and they are not interchangeable. Some only sort things into buckets. Some only predict what you might like. Some generate new text and images. A few can take actions in the real world, like sending an email or updating a calendar.
These differences matter in daily life. A recommendation system can only rank things that already exist. A generative model can produce something new, and that also means it can produce something wrong. An agent can take actions for you, so it needs permissions the others never do. When you know which kind you are looking at, you know what to expect from it and what to double check.
The two questions that sort them
When you meet a new AI feature, ask what it can see and what it can change. A chatbot sees your message and changes a text box. An agent might see your inbox and change your calendar. Those are very different levels of trust, and the rest of this series builds up to how that difference works.
These two questions also tell you what a product simply cannot do. A chatbot with no access to your calendar cannot double-book you, no matter how badly it answers. An agent that can read your inbox deserves the same scrutiny you would give a new assistant in their first week. The more a system can see and change, the more careful you should be with it, and that has nothing to do with how smart it is.