Categories of AI

This entry explains the categories of AI.

It is no exaggeration to say that not a day goes by without hearing the word “AI.” Almost every day brings news of a new AI service, or of yet another way to boost productivity with AI.

In this environment, “AI” has become a catch-all term — a word that means different things to different people. Let us get on the same page.

Specifically, this entry addresses the following questions:

  • What is “AI”?
    • What is an “AI agent”?
  • What kinds of AI are there?
  • Which AI should you start with?

Since this entry starts from the very basics, feel free to skip the parts you already know.

Let us begin.

toc

What Is AI?

By dictionary definition, “AI” stands for “artificial intelligence.” Think of it as a general term for the technology that reproduces human intelligence on a computer.

Keyword

AI / artificial intelligence (dictionary definition): a general term for the technology that reproduces human intelligence on a computer

That said, as of 2026, defining “AI” simply as services (apps and software) built on generative AI technology, such as ChatGPT or Gemini, works in nearly all situations.

What does this mean?

What Is “Generative AI”?

“Generative AI,” as the name suggests, refers to artificial intelligence that can generate something, such as text or images.

Although generative AI itself has existed for a long time, its performance was long nowhere near human intelligence. It was once mocked as “artificial idiocy.” For that reason, conventional AI focused on tasks like classification and prediction, and generation received little attention.

In the 2010s, however, several technical breakthroughs (Deep Learning, the Transformer architecture, and others) took place, and the performance of generative AI began to improve dramatically. Those advances came to fruition in the 2020s, symbolized by the release of ChatGPT in 2022. The AI boom that followed needs no introduction.

As a result, when we use the word “AI” today, we almost always mean a specific service built on generative AI technology. With that in mind, let us adopt the following as the practical definition of AI.

Keyword

AI (practical definition): a service built on generative AI technology (e.g., ChatGPT, Gemini)

Since this definition is more useful for anyone who is not an AI researcher, this site uses “AI” in this sense.

Categories of AI

What kinds of AI are there, specifically? Here is an overview.

The AI LandscapeKey Services & FeaturesWhat It IsMain Uses• ChatGPT• Gemini• Claude• Others: Grok, Llama, etc.• Artificial intelligence itself(Think of it as “a very smart person whoknows most publicly available information”) ‒ This intelligence is called a “model”• As a teacher ‒ Research and information gathering ‒ Learning and study support• As a worker ‒ Generating ideas ‒ Writing and translating• As a discussion or conversation partner ‒ Bouncing ideas off it, brainstorming ‒ Casual conversationAI Agents• Claude Code• Codex• ChatGPT agent• Gemini Spark• An LLM given permission to act outside the chat ‒ Plans and carries out tasks on its own: fileoperations, running code, browsing, etc. ‒ Currently offered as a feature of LLMs• Programming and app development• Automating routine tasks• Speeding up document creation• Web research and tasks done on yourbehalf• Perplexity (research)• NotebookLM (reading yourdocuments)• Microsoft 365 Copilot (Office)• Cursor (programming)• A service that tailors an existing LLM to a specific fieldor use ‒ Providers usually run the LLMs above behind thescenes rather than building their own ‒ What sets them apart: the UI and the knowledge(data) fed to the LLM• Research and reading documents• Help with work inside business apps• Industry-specific uses: medicine, law,education, etc.Image• Midjourney• Stable DiffusionVideo• Veo• Seedance• RunwayMusic & Voice• Suno• Udio• ElevenLabs (voice)• Services with their own non-LLM models, specialized in generating content other than text ‒ ChatGPT and Gemini can also create images and videos, but dedicated generation models runbehind the scenes• Just describe what you want in words, and you get decent results quickly ‒ What you “cannot do” suddenly becomes what you “can do”Specialized AILLMs /General-Purpose AICreative AI

For clarity, I have kept the number of specific services to a minimum. Since AI itself can list as many as you like, focus here on the framework of categories. Once you understand that, you will be able to choose which AI to focus on.

I have organized AI into the following framework:

  • LLMs / general-purpose AI
    • AI agents
  • Specialized AI
  • Creative AI
    • Image generation
    • Video generation
    • Music and voice generation

Treat all of these labels as my own coinages; the vocabulary in this area is still far from settled.

Let us go through them in order.

Categories of AI #1: LLMs / General-Purpose AI

First comes the most fundamental kind of AI. Let us call AI that can be used for many different purposes, like ChatGPT, general-purpose AI. The three best-known examples are these:

  1. ChatGPT
  2. Gemini
  3. Claude

You interact with these AIs through text-based chat. If you have never tried any of them, open one of the links above and try it now.

After using one for a little while, you will notice the strange sensation that there is another person on the other side of the screen. This, truly, is “artificial intelligence.”

Keyword

General-purpose AI: AI that can be used for many different purposes

LLMs

Since all general-purpose AIs are built on a technology called the Large Language Model, they are also commonly referred to as LLMs. Learn this term alongside “general-purpose AI.”

Keyword

LLM (Large Language Model): the AI technology behind general-purpose AI. For practical purposes, it can be treated as a synonym for “general-purpose AI”

From here on, this entry mainly uses “LLM” to mean general-purpose AI (since “LLM” is the more appropriate term in technical contexts, and using “AI” everywhere makes the writing harder to follow).

What LLMs Can Do

A useful mental image for an LLM is a very smart person (or brain) who knows almost everything that is publicly available. This person supports you tirelessly. Almost every concrete use falls into one of these three categories:

  • Use it as a teacher
    • Gathering information and looking things up
    • Supporting your learning and studies
    • Correcting your writing
  • Have it do work for you
    • Generating ideas
    • Writing and translating text
  • Use it as a discussion or conversation partner
    • Bouncing ideas off it, brainstorming
    • Casual conversation

It is hard to imagine anyone for whom none of these is useful. Since you are likely to find some value once you try it, I recommend starting with an LLM.

The only differences between the free and paid plans are how smart the model is and how often you can use it. Starting with the free plan is fine.

Which LLM Should You Use?

There is no need to agonize over whether to use ChatGPT, Gemini, or Claude. Since each company adds new features every few months, “which one is best” keeps changing. The following approach is enough:

  • If you already pay for one of them, use that one
  • If not, try all three for free and pick the one that works best for what you do

AI Agents

The term “AI agent” comes up often in the context of LLMs. Let us sort it out.

Although the term is currently used in many different senses, on this site it means “an LLM given permission to act in environments outside the chat (on your desktop or in a browser).”

Keyword

AI agent: an LLM given permission to act in environments outside the chat

To hand complex work over to an LLM, you need to give it permission to act outside the chat. For example, to have it build a program made up of many files, you must give the LLM permission to read, write, and create files. Within a chat, the most it can do is output code that fits in a single file (a macro, for example).

In other words, “handing complex work to an LLM” basically means “turning the LLM into an agent.” Of the uses listed under “What LLMs Can Do,” using it as a teacher or as a discussion or conversation partner can be done entirely within a chat. Having it do work for you, on the other hand, increasingly requires an agent as the work grows in scale.

For example, I used Claude Code to build a system that automates tedious tasks. Although I am not new to programming, I had never even touched the languages that system uses (Python and shell scripts). Even so, the system works just as I intended. We truly live in remarkable times.

Do AI Agents Have to Keep Working on Their Own?

Incidentally, the common definition of “AI agent” often includes operating autonomously over long periods. The image is that once a human gives the initial instructions, the AI keeps working on its own for hours.

In my opinion, however, this definition goes too far for now.

For an AI to keep operating autonomously, one of the following conditions must be met:

  1. The task can be self-verified: the AI can check for itself whether it is heading in the right direction
  2. There is a large volume of tasks, each with a low cost of failure: even without self-verification, humans can check the results in bulk afterward

Since only organizations above a certain size have tasks that fit #2, we can treat that case as an exception.

That leaves #1, and right now, the area most likely to meet this condition is coding. Once tests are in place, the AI can check for itself whether the code works correctly.

That said, not all coding can be left to AI for hours. Although I gave AI some fairly heavy tasks while developing a game with it, it never once worked autonomously for more than an hour. Since tests cannot tell whether a game is fun to play or looks the way I intended, I had to check those things myself. Long autonomous runs work only for coding where correctness can be judged mechanically, such as large-scale rewrites or fixes aimed at passing a test suite.

In other words, if we include long-running autonomous operation in the definition of “AI agent,” then “AI agent” effectively becomes “having AI write code that tests can verify.” There would be little point in using a new term for that.

Of course, although AI that keeps working autonomously may eventually appear in areas other than coding, there are not many areas where “what is right” can be determined mechanically. That is why I think it is better not to give “AI agent” too narrow a meaning.

Prediction: LLM = AI Agent

The line between “chat AI” and “AI agents” will most likely fade over time, because it will become normal for AI to have permission to act in external environments. In fact, ChatGPT and Gemini have already begun building agent features into their apps.

As you will see if you try one, AI agents such as Claude Code are, in terms of capability, a superset of regular chat mode (though they come at a cost: they are slower, and they carry a risk of mistaken actions). There is almost nothing that only chat can do. On the other hand, there are many things that cannot be done without granting permission to act. Coding, mentioned above, is the prime example.

That is a brief overview of LLMs.

Categories of AI #2: Specialized AI

The AI LandscapeKey Services & FeaturesWhat It IsMain Uses• ChatGPT• Gemini• Claude• Others: Grok, Llama, etc.• Artificial intelligence itself(Think of it as “a very smart person whoknows most publicly available information”) ‒ This intelligence is called a “model”• As a teacher ‒ Research and information gathering ‒ Learning and study support• As a worker ‒ Generating ideas ‒ Writing and translating• As a discussion or conversation partner ‒ Bouncing ideas off it, brainstorming ‒ Casual conversationAI Agents• Claude Code• Codex• ChatGPT agent• Gemini Spark• An LLM given permission to act outside the chat ‒ Plans and carries out tasks on its own: fileoperations, running code, browsing, etc. ‒ Currently offered as a feature of LLMs• Programming and app development• Automating routine tasks• Speeding up document creation• Web research and tasks done on yourbehalf• Perplexity (research)• NotebookLM (reading yourdocuments)• Microsoft 365 Copilot (Office)• Cursor (programming)• A service that tailors an existing LLM to a specific fieldor use ‒ Providers usually run the LLMs above behind thescenes rather than building their own ‒ What sets them apart: the UI and the knowledge(data) fed to the LLM• Research and reading documents• Help with work inside business apps• Industry-specific uses: medicine, law,education, etc.Image• Midjourney• Stable DiffusionVideo• Veo• Seedance• RunwayMusic & Voice• Suno• Udio• ElevenLabs (voice)• Services with their own non-LLM models, specialized in generating content other than text ‒ ChatGPT and Gemini can also create images and videos, but dedicated generation models runbehind the scenes• Just describe what you want in words, and you get decent results quickly ‒ What you “cannot do” suddenly becomes what you “can do”Specialized AILLMs /General-Purpose AICreative AI

Next, let us call services that tailor an existing LLM to a specific purpose or field specialized AI.

In most cases, the providers of specialized AI do not have their own LLM. Instead, they run LLMs such as ChatGPT or Claude behind the scenes and set themselves apart in two ways:

  • UI: the screens and controls are designed to make specific tasks easier
  • Knowledge: specific data is fed to the LLM
    • Web search results, documents the user provides, information from a specialized field, etc.

For example, Perplexity specializes in research, NotebookLM in reading documents you provide, and Cursor in programming. There are also many specialized AIs for particular industries, such as medicine, law, and education. See the table for specific examples.

Keyword

Specialized AI: a service that tailors an existing LLM to a specific purpose or field

Since the core is an LLM after all, a general-purpose LLM can do much the same to some extent. If you use AI for a particular purpose often, however, specialized AI is likely to take less effort and give better results. It is worth checking once whether there is a specialized AI for the things you use AI for most.

Categories of AI #3: Creative AI

The AI LandscapeKey Services & FeaturesWhat It IsMain Uses• ChatGPT• Gemini• Claude• Others: Grok, Llama, etc.• Artificial intelligence itself(Think of it as “a very smart person whoknows most publicly available information”) ‒ This intelligence is called a “model”• As a teacher ‒ Research and information gathering ‒ Learning and study support• As a worker ‒ Generating ideas ‒ Writing and translating• As a discussion or conversation partner ‒ Bouncing ideas off it, brainstorming ‒ Casual conversationAI Agents• Claude Code• Codex• ChatGPT agent• Gemini Spark• An LLM given permission to act outside the chat ‒ Plans and carries out tasks on its own: fileoperations, running code, browsing, etc. ‒ Currently offered as a feature of LLMs• Programming and app development• Automating routine tasks• Speeding up document creation• Web research and tasks done on yourbehalf• Perplexity (research)• NotebookLM (reading yourdocuments)• Microsoft 365 Copilot (Office)• Cursor (programming)• A service that tailors an existing LLM to a specific fieldor use ‒ Providers usually run the LLMs above behind thescenes rather than building their own ‒ What sets them apart: the UI and the knowledge(data) fed to the LLM• Research and reading documents• Help with work inside business apps• Industry-specific uses: medicine, law,education, etc.Image• Midjourney• Stable DiffusionVideo• Veo• Seedance• RunwayMusic & Voice• Suno• Udio• ElevenLabs (voice)• Services with their own non-LLM models, specialized in generating content other than text ‒ ChatGPT and Gemini can also create images and videos, but dedicated generation models runbehind the scenes• Just describe what you want in words, and you get decent results quickly ‒ What you “cannot do” suddenly becomes what you “can do”Specialized AILLMs /General-Purpose AICreative AI

Finally, let us call AI specialized in generating content other than text creative AI. Its main areas are these three:

  1. Images
  2. Video
  3. Music and voice

There are also AIs that generate 3D models, among others.

Keyword

Creative AI: AI with its own non-LLM model, specialized in generating content other than text

Creative AI feels less like “artificial intelligence” and more like an extraordinarily useful tool. In fact, image and video generation AI now lets anyone create reasonably good images and videos.

The Line Is Getting Blurry

However, AI agents may be able to do what creative AI does.

For example, music generation AI such as Suno and Udio can compose songs. But you can also make music by giving an AI agent instruments (sound libraries). In fact, I had Claude Code make background music using Python and the instrument sounds that come with GarageBand.

In other words, if an LLM already knows how to make something good, connecting an AI agent to software or an environment that can produce it lets the agent make it for you. Since an LLM knows which chord progressions and melodies tend to work, music comes out once you give it the tools.

My personal prediction is that connecting agents to tools will become the mainstream approach (it adds no extra cost, and you can fine-tune the results). If you are trying out AI agents, it is worth checking what they can connect to and what they can make.

Try Making Creative Work Too

Whether you use creative AI or an agent, what you should make depends on the person. After all, images and music are not needed as often as text.

That said, partly as a note to myself, I think the better attitude is to try making everything at least once.

The reason is that AI suddenly turns “cannot do” into “can do.” Specifically, things like the following really happen:

  • You become an illustrator overnight
  • You become a filmmaker overnight
  • You become a musician overnight

Of course, none of this means you will reach a professional level instantly. Still, there is no doubt that you can now get to “can do” (producing actual work) at a speed that was unthinkable before.

In this situation, the reasoning “that has nothing to do with me, so I will pass” no longer holds. If you can learn to do something right away, you can decide whether you need it after you have learned it.

A good first step is to try the image generation feature in ChatGPT or Gemini. Although these are LLMs by category, a separate generation model runs behind them. If they stop being enough for you, I also recommend trying a dedicated service.

That covers what AI is and how it can be categorized. Next, I will give an overview of what AI can and cannot do. Please continue to the following entry.

All AI-related entries are collected below.