Just finished building my “Vulgar Snake” game with aider and Deepseekv3 - and the results are staggering compared to my previous experience building “Snake by Michael Bay” with RooCode and Gemini.
Token Efficiency Breakdown:
- aider + Deepseekv3 (Vulgar Snake): ~31k tokens total
- RooCode + Gemini (Michael Bay Snake): ~3M tokens (yes, million!)
That’s nearly 100x more efficient token usage with aider! Here’s why it worked so well:
- Precise, focused changes - aider only modifies what I ask for, no wasteful regenerations
- Minimal context needed - Deepseekv3 understands the existing code structure with minimal prompting
- No boilerplate bloat - aider keeps the code lean and focused on functionality
Snake by Michael Bay (RooCode) features:
- Explosions (with particle effects)
- Screen shake
- Slow motion effects
- All in just 6 clean JavaScript files
Vulgar Snake (aider) features:
- NSFW language and humor
- Dynamic difficulty scaling
- Minimalist design
- Even more efficient codebase
Biggest lesson? You don’t need massive context windows or expensive models to build great stuff. Focused tooling + smart prompting beats brute force every time.
Check out both games here:
- Snake by Michael Bay (RooCode)
- Vulgar Snake (aider)