Overview
A former AI coding skeptic documents their experience with modern coding agents, discovering that recent models can handle months-long programming projects that would challenge experienced developers. The post traces a progression from simple scrapers to attempting to port Python’s scikit-learn library to Rust using AI agents.
The Breakdown
- Simple automation evolved to complex systems - starting with YouTube metadata scrapers and progressing to porting entire machine learning libraries between programming languages
- Modern coding agents handle multi-month projects - the author attempted to recreate scikit-learn in Rust, implementing fast algorithms that outperform the original Python implementations
- Model capability jumped dramatically in late 2024 - Opus 4.6 and Codex 5.3 represent an order of magnitude improvement over models released just months earlier
- Skeptics face a communication problem - describing these capabilities sounds like hype but reflects genuine surprise at solving previously intractable coding challenges