Cursor

Software is created in a special writing program known as an editor. There, developers type the instructions that later become an app or a website. Cursor is such an editor, but with one crucial addition. Built into it is an AI that can itself read, extend, and rewrite these instructions. You can tell it in plain language: “Add a login here.” It then searches through the existing files and suggests the necessary changes. Cursor comes from the US company Anysphere and became well known starting in 2023.

Why developers are suddenly talking about it

For a long time, programming largely consisted of tedious work. Much of it is routine: shuffling data from A to B, building forms, catching error messages. It is exactly this routine that an AI can now handle remarkably well. Those who use Cursor type less and check more. The role shifts from writer to reviewer.

Economically, this is a big deal, because developer hours are expensive. Anysphere therefore grew unusually fast and was valued at double-digit billions in funding rounds. Major providers such as Microsoft with GitHub Copilot or Anthropic with Claude Code are pushing into the same market. Cursor is seen as an example of how a small team can build a highly valuable product using third-party AI models.

But there is also criticism. AI-generated code often looks correct but doesn’t work or contains security vulnerabilities. Anyone who accepts the suggestions without checking them merely pushes the problem further down the line. Experts therefore say that Cursor is less useful for beginners than for experienced developers who can immediately spot mistakes.

What happens in the background while typing

Cursor itself does not have its own large AI model for all tasks. It sends requests to third-party models, for instance from OpenAI or Anthropic, and processes the responses. The actual trick lies in which information is sent along. Experts call this the context: everything the model needs to know about the project.

A project can consist of thousands of files, and not all of them fit into a single request. Cursor therefore searches through the project and selects the presumably relevant parts. It also keeps track of which file is currently open and what was most recently changed. An analogy: instead of handing the AI the entire library, Cursor lays out the three matching books already opened for it.

On top of this there is a so-called agent mode. Here the AI doesn’t just work through a single suggestion but carries out several steps one after another. It changes a file, runs tests, reads the error message, and makes corrections. All changes appear highlighted in color and must be confirmed or discarded individually. This confirmation step is the most important safeguard against unnoticed errors.

Who works with it and what it costs

Cursor runs on Windows, macOS, and Linux and is based on the free editor VS Code. That’s why people switching over find their way around quickly, since the operation and extensions are largely the same. There is a free version with limited usage as well as subscriptions for individuals and companies. In businesses, the key question is above all whether one’s own code ends up on third-party servers.

In the news, Cursor mostly comes up in three contexts. First, in funding rounds and valuations, since the company is among the fastest-growing software providers. Second, in the debate over whether AI is changing or replacing programming jobs. Third, in disputes over pricing, since every request to a large model costs real money.

Anyone wanting to get started themselves doesn’t need a degree. School projects, small websites, or games can be built noticeably faster with it. A common misconception, however, is that this means you no longer need to learn programming. Without a basic understanding, you won’t notice when the AI is producing nonsense.