An AI app is an application whose core behavior runs through a model instead of hand-written rules. You hand it open-ended input — a sentence, an image, a document, a question — and it returns a judgment: a classification, a summary, a generated draft, a decision. The logic that used to live in code you wrote now lives in a model you prompt, and that one shift is what puts the "AI" in the app.
That's a different thing from software with a machine-learning feature bolted on. A recommendation row or a spam filter uses a model, but the product around it is ordinary software. An AI app is one where the model is the product — take the model out and there's no app left. This lesson maps the categories, shows you real examples, and names the one structural property they all share, so the rest of this course has a clear thing to build.
What makes an app an AI app
The line isn't "does it use AI somewhere." Almost everything does now. The line is whether the model sits in the critical path of what the user came for. When the model produces the thing the user wanted — the answer, the draft, the extraction, the decision — you have an AI app. When the model just ranks or nudges around a conventional product, you have software with an AI feature. The difference matters because everything downstream — how you test it, how it fails, how you ship it — follows from the model being load-bearing.
The main categories of AI apps
Most AI apps fall into a handful of shapes. You'll recognize products you already use in each row, and you'll build things that fit these shapes in the lessons ahead.
| Category | What it does | Example you know |
|---|---|---|
| Assistant / chatbot | Holds a conversation and answers open-ended questions | ChatGPT, customer-support bots |
| Copilot | Suggests or drafts inside a tool you're already using | GitHub Copilot, email draft suggestions |
| Extraction / automation | Pulls structured data out of messy input and acts on it | Invoice parsers, resume screeners |
| Generative product | Produces new content from a prompt | Image generators, marketing-copy tools |
| Classification / routing | Labels or routes input by intent or content | Ticket triage, content moderation |
Why AI apps need a different kind of engineering
Here's the property every AI app shares: a probabilistic component sits in the request path. A normal function returns the same output for the same input; a model doesn't have to. Feed it the same request twice and you can get two different answers, and the tenth user will send something you never imagined. That single fact is why you can't build an AI app the way you build a form-and-database app — and why "AI engineering" exists as its own discipline. You still test, version, and ship, but you're doing it around a component that behaves like a person, not a pure function.
That's the thread this course pulls. Next, you'll meet the person who does this work day to day — read "What does an AI engineer do?" to see what an AI engineer actually builds, and the skills the role demands, before you build your first pipeline yourself.