Do people have thoughts on Conductor vs Warp vs just running Claude Code in iTerm2?
Just tried Conductor and feeling a little unsatisfied but I can’t put my finger on why.
I’m pretty pilled on specs being part of the codebase.
Makes it way easier for LLMs to understand features and they’re way better at keeping the spec up to date as requirements change.
Latest @windsurf_ai vs @cursor_ai test, both using Claude 3.7 Sonnet: A race track.
Prompt: Let's build a simple game where a user drives a car around an elliptical track using html, css, and javascript with an express server to run the page. They will use their keyboard with the keys "W,A,S,D" to control direction. Every lap registers a new lap time and you see your current lap time, latest lap time, and fastest lap time on the screen.
Similarities:
- Designs are identical
- Keyboard functionality works
- Both show instructions on the page
Windsurf
- Cannot turn unless you are moving
- has track boundaries, but they are invisible
- track boundaries are also very tight which means you have to be very precise with your driving
- Timer doesn't work
Cursor
- Can turn even if you aren't moving (easier to drive)
- No track boundaries
- Precise marker to track start and end lap times which works, but because there are no track boundaries, it is easy to cheat
Next test will be a mobile game built using Expo.
Follow up with the next test: Flappy Bird.
Windsurf was significantly better than Cursor here.
Windsurf: 90% success rate
- Design was decent. Bird looked like a bird and pipes had some definition.
- Collision detection was pretty good, but very conservative. If you were close to the pipe, it considered it game over.
- Up and down motion was a little jumpy. Could have been smoother.
- Functional scoring.
Cursor: 50% of the way there
- Design was very basic. The bird was just a ball and pipes were simply green cylinders.
- The pipes don't even appear until about two seconds into the game.
- Collision detection was completely broken. Even if you were to completely avoid a pipe, the game would end.
- Scoring did not work.
- Up and down motion and gravity in general was way smoother than Windsurf though.
Been doing a little vibe coding of games for fun to compare Cursor and Windsurf.
Been using the prompt, "Build x game in html, css, js using the next.js framework," and seeing how the output functions.
Games tried so far:
1. Tic Tac Toe
2. Chess
Both handled Tic Tac Toe with ease. No additional input from me needed.
Chess was different. Both Cursor and Windsurf built a chess game. But using my sole prompt, Windsurf built out a fully featured chess game with move validation, check and checkmate status, move history, captured pieces, styling, etc.
Cursor on the other hand, built the a barebones chess game and then outlined for me the next few milestones which would get me to feature parity with the Windsurf version. Once you worked through the milestones, the two versions were the same.
Going to keep pushing forward here. Next up is Flappybird (will not use next.js for this one).
I find it unnatural to cast any thought I have. And I end up swinging to the other end of the spectrum and only casting fully polished thoughts. I need help finding more of a balance to post more.
Today was just one of those days. Son woke up early and was in a mood. My wife and I are exhausted.
But he’s ending the day with 4 teeth so hopefully tomorrow is much better!
