AI coding means using an AI model to write, check and change the code of software. The model is a large language model (LLM), the kind of AI behind chatbots, trained on huge amounts of text and public code. Since 2021, AI coding has moved through four stages. It began by finishing the line a developer was typing. Then it wrote a whole function from a short comment. Today an agent can take a whole task and return a finished change. At the far end, a builder makes a complete app for someone who cannot write code at all, such as a bakery owner who wants an ordering app and has no engineers.
First, a few words. Code is the set of instructions, written in a programming language, that makes an app work. A team keeps all its code, with its full history, in a repository. A new change is first offered as a pull request, a proposed change that other developers review before it joins the main code.
Take, for example, a product team at a lending app in Bengaluru. It wants a new feature: a screen that shows each borrower the EMIs left on a loan, and lets the borrower prepay. The product manager first describes the screen in plain English to an AI app builder. Within an hour she has a rough working version, called a prototype, to show five customers.
The engineers then build the real feature inside the main app. They write code in a code editor, the program in which developers type code. An AI coding assistant in the editor suggests each next line as they type. One smaller task, "show the prepayment charge before the borrower confirms", goes to an AI coding agent. The agent reads the repository, writes the change, runs the tests and opens a pull request by itself.
Every pull request, from a person or from the agent, is read first by AI code review. Within a few minutes it posts comments, such as: "This line divides by the number of EMIs left, which can be zero." It also writes a short summary for the human reviewer. A senior engineer then reads the change and the comments, and approves it. Only then does the CI/CD pipeline build the app, run all its tests and deploy it. To deploy means to install the new version on the servers that customers' phones connect to. A server is a computer in a data centre that runs the backend of an app: the part that stores data and does the calculations.
The last product works inside the finished app. With generative UI, a borrower can type "What if I prepay ₹50,000 now?" A model then puts together a small screen for that one question, with a table and a slider, instead of replying with a paragraph.
What does the AI not do? It does not decide what to build, and it is not responsible for the result. Its code can look correct and still be wrong. So automated tests, small programs that use the app as a person would and check the result, and a human reviewer check every change before customers see it. The idea behind the agent is explained in AI coding agents.