Cursor in practice
Cursor is a code editor designed around an AI agent. You write in it as in a classic editor, with help all the way: the agent explains a file, proposes a plan before touching the code, then applies a change across several files when asked. Every change is shown before it is accepted. For a company, it mainly serves to go faster on the repetitive code of a business application or an MVP.
Getting started with Cursor
- Install the editor from the vendor’s website and open your project folder.
- Start with questions: ask the agent to explain how the project is organized before having it write anything.
- For a change that touches several files, ask for a plan first, read it, then let the agent apply it.
- Review each proposed change and run the tests before approving.
Project rules, the setting that changes everything
An agent works better when it knows the team’s conventions. In Cursor, they are written as project rules, stored in the repository’s .cursor/rules folder, or in a simple AGENTS.md file. That is where you note naming conventions, the commands to use, the chosen architecture and the known pitfalls. These rules are versioned with the code: the whole team works with the same instructions.
What to check before adopting it
To work, the editor sends extracts of your code to the model. Turn on privacy mode, check in your contract that the code is not used to train models, and keep passwords and access keys out of the repository. Also keep a human review on every pull request: the tool writes fast, it is not accountable for the result. Our Cursor or Claude Code comparison details what separates the two approaches.
