An AI agent coding skeptic tries AI agent coding, in excessive detail

2026-02-27 · Source: Max Woolf's Blog · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Software Development & Engineering, Data Science & Analytics · Depth: Advanced, extended

Summary

This article details an experienced data scientist's evolving perspective on AI agent coding, shifting from skepticism to optimism following significant improvements in models like Claude Opus 4.5 and GPT-5.3-Codex. The author initially found agents unpredictable and expensive but later used them for practical tasks, including creating a Python wrapper for Google's Nano Banana generative AI and implementing a `Grid` class for image manipulation. A key finding was the effectiveness of `AGENTS.md` files for controlling agent behavior and code quality. The author successfully used agents to develop several Rust projects with Python bindings, such as `icon-to-image`, a terminal-based MIDI composer `miditui`, and a physics simulator `ballin`. Critically, agents achieved 2-100x speedups for various machine learning algorithms like UMAP, HDBSCAN, and GBDT, even matching or exceeding highly optimized libraries like NumPy and XGBoost, by iteratively optimizing Rust code.

Key takeaway

For Machine Learning Engineers and Data Scientists seeking to optimize computationally intensive algorithms, consider integrating modern AI agents like Claude Opus 4.5 or GPT-5.3-Codex into your development workflow. By leveraging `AGENTS.md` files and iterative optimization prompts, you can achieve substantial performance improvements (2-100x) in Rust implementations with Python bindings, potentially surpassing existing highly optimized libraries. This approach offers a path to developing faster, more robust data science tools, even for complex tasks like UMAP or HDBSCAN.

Key insights

Modern AI agents, guided by explicit `AGENTS.md` files, can generate highly optimized, performant code across multiple languages.

Principles

• Explicit agent configuration via `AGENTS.md` is crucial for quality.
• Iterative optimization by agents yields significant performance gains.
• Combining agents can achieve superior optimization results.

Method

Define agent behavior with an `AGENTS.md` file, provide detailed prompts, and iteratively optimize generated code using benchmarks and accuracy checks, potentially chaining different LLMs for enhanced performance.

In practice

• Use `AGENTS.md` for consistent code quality and style.
• Employ `pyo3` to bridge Rust's speed with Python's ecosystem.
• Benchmark agent-generated code for quantifiable performance improvements.

Topics

• AI Agent Coding
• LLM Code Optimization
• Rust Development
• Machine Learning Algorithms
• AGENTS.md Configuration

Code references

• minimaxir/gemimg
• features/copilot
• minimaxir/youtube-video-scraper
• minimaxir/youtube_scraper_opus
• huggingface/tokenizers

Best for: Machine Learning Engineer, Data Scientist, Software Engineer

Related on AIssential

• Writing a Good AGENTS.md — Effective `AGENTS.md` files are concise, human-written, and focus on essential, non-redundant information.
• A quick chat with Ines & Matt — Agentic systems can generate "software 2.0" by encoding behavior in data examples, not just explicit code.
• Your Coding Agent Should Do AI System Engineering — Ben Burtenshaw, Hugging Face — Coding agents can perform complex AI systems engineering tasks, from kernel optimization to automated research.
• Applying an Agentic Coding Tool for Improving Published Algorithm Implementations — Agentic coding tools can significantly improve published algorithm implementations across diverse domains within a single workday.
• Coder Agents Enable Running AI Coding Workflows on Self-Hosted Infrastructure — Self-hosting AI coding agents provides control over code, data, and execution, reducing vendor lock-in.
• Can Coding Agents Reproduce Findings in Computational Materials Science? — LLM-based coding agents currently struggle to reproduce computational materials science claims due to procedural and execution challenges.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Max Woolf's Blog.