Overview
Max Woolf, a former AI coding skeptic, documents his journey testing coding agents on increasingly ambitious projects. His experience reveals that modern AI agents can tackle complex, multi-month programming tasks that would challenge experienced developers, fundamentally changing what’s possible in automated software development.
The Breakdown
- Started with simple projects like YouTube metadata scrapers and progressively scaled to more complex challenges to test AI agent capabilities
- Attempted an extremely ambitious project: porting Python’s scikit-learn machine learning library to Rust using AI agents, including fast implementations of algorithms like logistic regression and k-means clustering
- Discovered that recent models (Opus 4.6/Codex 5.3) represent an order of magnitude improvement over previous coding LLMs released just months earlier, despite the difficulty of conveying this without sounding like hype
- Found that AI agents consistently handle complex tasks that would take experienced developers months to complete, breaking the traditional boundaries of what automated coding can accomplish