Cursor AI Review
A fork of the beloved VSCode with a supercharged agent to speed through any development project, but with a few drawbacks and shortcut mayhem.
I’ve been a VSCode user for as long as I can remember, and have been using CoPilot for the past year. However, the rise of AI has left us spoilt with choices, but the Cursor AI IDE has been my choice for the past 6-months.
Background
Development of Cursor started back in 2023 by their parent company AnySphere, however it’s only recently entered a stable build in past month (June 2025) with it’s 1.0 build. Therefore, I’ve unknowningly been a beta tester for the past 5-months, as I started using it back in February.
Cursor has 3x modes ‘Agent’, ‘Ask’ and ‘Manual’, I’ve been predominantly using ‘Agent’ which has been let loose and changes my files directly. Utilising multiple LLM’s from OpenAI (GPT4.1 etc), Antropic (Claude-4-Sonnet) and Google (Gemini), there is various options depending on the task at hand.
The parent company AnySphere isn’t holding back they’ve recently bought the AI House Koala in an attempt to capture top-talent and compete with the big players.
User Interface
As mentioned, Cursor is just a fork of VSCode, the UI is familiar and creature comforts in the form of the git tab and extensions are still all active. However, my muscle memory has been causing a headache, as with more features comes more shortcuts and in the case of Cursor some shortcuts overlap. Other than that though, it’s visual appeal is comforting and easy to pickup for us VSCoders.
Performance
I will precipt this with the fact I have been paying $20 (14.96 for the fellow Brits) for the Pro version. As a result, I’ve had the fastest and smartest LLM’s at hand and have never encountered any rate limits. The performance and accuracy has been generally good, feed it good prompts and get good results. A major benefit of Cursor is it’s awareness of the entire codebase, that is where it shines, code completions and agent mode are accurate nearly everytime and it’s testament to it’s memory and ability to read, understand and code based on what’s already there. I have recently started using the MCP documentation integration, which feeds the LLMs the cutting edge information of the tools and languages I’m using, which has been vital to this vary blog which is built on an Astro framework.