AI coding agent one-shot challenge with Amp, Claude Code and Gemini
9 months ago
- #side projects
- #productivity
- #AI coding agents
- The author discusses their struggle with managing numerous side projects on GitHub due to limited time.
- They explore AI coding agents to increase productivity, aiming for a 10x or 100x throughput improvement.
- Three AI coding agents (Amp, Claude Code, Gemini) were tested with a one-shot challenge to build a web application.
- Amp had issues with configuration, routing, and build errors but provided a good project structure.
- Claude Code delivered a working web UI but didn't follow the requested code structure and lacked CSS.
- Gemini attempted to use Shadcn UI but failed to create a functional web app and had multiple configuration issues.
- Claude Code was the winner for delivering a working UI, but all agents had significant shortcomings.
- The author plans to refine prompts and test more agents in future rounds for better results.