AI is not one thing, it is a few kinds of tool wearing a hundred different logos. This piece maps the field so it stops being a blur, and two plain questions place almost all of it.
Open any app store and it looks like a new AI drops every day, each one swearing it will change your life. It is less crowded than it looks. Under all the logos sit only a few real kinds, and every one of them runs on the idea from the earlier piece, a system that pulls patterns out of past examples. What splits them apart is not that idea. What do you feed in and what comes back out? Can it only talk, or can it also go and do something? Those two questions place the whole field.
This is an overview of the three most important kinds of AI models right now. After this, the series goes deeper into each one.
Language models
This is the chatbot, the one you already know. You give it text and it gives you text back, and its entire world is language. That makes it strong at anything built from words, explaining a topic, drafting a message, summarizing a mess, talking a problem through, and it means it cannot touch a single thing outside the conversation. It writes you the recipe. It never cooks.
Three names own this space, ChatGPT, Claude and Gemini. ChatGPT is the default because it opened the door first and let anyone in free, so it became the household word the way Google did for search. Gemini rides Google’s scale and sits baked into the search bar, the phone and the inbox you already live in. Claude is the one picked for the quality of what comes out, the go-to for a lot of people who write and code for a living. The lead trades hands every few months, so the winner this week matters less than what keeps all three on top, the most training behind them, the most polish over the top and the shortest path to just start typing.
Image generators
Feed this one words and it hands back a picture. Same learning, pointed at a different job. You describe something and it paints it out of nothing in seconds, and that same idea now reaches into video and music too. The tell is the swap, words go in and something to look at comes out, not something to read. This is also the corner carrying the most anger from the last piece, because these models learned from artists’ work that got scraped without asking or paying, and that fight is still live, in courtrooms as much as in comment sections.
Agents
This one sits a level deeper. An agent is a language model handed hands and a goal. Instead of answering and stopping, it takes steps, uses other software, checks its own work and keeps going until the task is done. The gap from a chatbot is the whole point. A language model tells you how to do something. An agent tries to do it for you. That makes it powerful and it makes it riskier, since it can run with its own mistakes, not only speak them.
A few more you’ll meet
The big three cover most of what gets called AI, but a handful of others are worth a line so you can spot them in the wild. The feeds on Netflix, Spotify and every social app are AI as well, the same pattern-learning aimed at your attention to pick what comes next. Voice tools ride on top of the talker, turning your speech into text and a voice back out, which is the assistant in your phone and the dictation in your keyboard. Translation apps run that same language trick pointed at another tongue. The quietest kind never opens a chat window at all, it sorts and predicts in the background, catching spam, flagging fraud and reading medical scans. None of these need a deep dive here. Clock them, and notice how much AI you were already leaning on without ever calling it that.
The short version
So here is the whole overview that matters for now, on one page. Peel the logos off and AI is a short stack of kinds. The language model you type to. The image generator that now makes video and sound as well. The agent that goes and does instead of only telling. Underneath all of it, the quiet feeds and filters you never think of as AI at all. Two questions place any of them, what goes in and what comes back, and whether it can only make something or also act on the world.
Now forget the names. The leaders will trade places, this month’s winner beaten by next month’s, and none of that is worth memorizing. The shape underneath does not move, and that is the part worth remembering. Each of the big three earns a fuller piece of its own down the line, but the light version comes now, because the map sticks better once you have seen what sits on it.
The next three pieces take the kinds up close, one apiece, starting with the one you touch every day. Now that you can place the language model on the map, we get in close and find out why it can sound so sure of itself and still be flat wrong.
Next up: A Closer Look at the Language Model →