Windsurf out-performs Cursor in my testing

I’ve been running an interesting experiment these past weeks, comparing two AI coding assistants side by side: Windsurf and Cursor. The setup was straightforward - Windsurf supporting my main Elixir/Phoenix application, while Cursor handled the React Native mobile development.

To make things interesting, I ran them on different models: Windsurf on 3.7-sonnet and Cursor on Gemini 2.5 pro. This gave me a clear picture of how they perform in real-world development scenarios.

What I’ve noticed is fascinating. Windsurf seems to be maturing rapidly, with consistent releases that improve its capabilities. The context retention has become remarkably stable, making it increasingly reliable for daily development work.

Cursor, on the other hand, has been showing some concerning behaviors. It’s become more prescriptive than productive - telling me what to do rather than actually helping me do it. The context losses and its reluctance to edit files directly have become significant friction points in my workflow. For now, I’ve decided to pause my Cursor usage, as Windsurf simply makes me considerably more productive.

As a side note, I’ve started exploring Aider, which shows tremendous promise. Haven’t had the chance to dive deep yet, but it’s definitely on my radar for further investigation.