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

2026-02-27 · Source: Simon Willison's Weblog · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Software Development & Engineering · Depth: Advanced, quick

Summary

Max Woolf, a self-proclaimed AI agent coding skeptic, details his experience with advanced coding agents like Opus 4.6 and Codex 5.3 through a series of increasingly complex projects. He began with simple tasks such as YouTube metadata scrapers and progressed to an ambitious undertaking: porting Python's scikit-learn library to Rust, creating a new crate named `rustlearn`. This project aims to implement fast versions of standard machine learning algorithms like logistic regression and k-means clustering, outperforming scikit-learn's implementations. Woolf emphasizes the significant improvement in these models, stating they are "an order of magnitude better" than previous coding LLMs, despite the difficulty in conveying this without sounding like hype. His work demonstrates the agents' capability to handle complex coding tasks that would typically require months for an experienced developer.

Key takeaway

For AI Engineers evaluating the current capabilities of coding LLMs, you should re-assess models like Opus 4.6 and Codex 5.3. Their ability to tackle multi-month coding projects, such as porting complex libraries or implementing machine learning algorithms, suggests a significant leap in agent proficiency. Consider integrating these agents into your development workflow for tasks previously deemed too complex for AI assistance, potentially accelerating project timelines.

Key insights

Advanced AI coding agents now handle complex tasks with surprising proficiency, challenging prior skepticism.

Principles

• AI agents excel at iterative code development.
• Complex tasks can be broken into agent-solvable steps.

Method

The author used a three-step pipeline for developing `rustlearn`: defining the problem, letting the agent generate code, and then refining/testing the output, even for complex algorithms.

In practice

• Use agents for porting existing libraries to new languages.
• Employ agents for implementing ML algorithms from scratch.

Topics

• AI Coding Agents
• Machine Learning Libraries
• Rust Programming
• scikit-learn
• Logistic Regression

Code references

• simonw/research

Best for: AI Engineer, Machine Learning Engineer, Software Engineer

Related on AIssential

• Closing the knowledge gap with agent skills — Agent skills effectively bridge LLM knowledge gaps for rapidly changing software development practices.
• An AI agent coding skeptic tries AI agent coding, in excessive detail — Modern AI agents, guided by explicit `AGENTS.md` files, can generate highly optimized, performant code across multiple languages.
• How AI Coding Agents Actually Understand Your Codebase — AI coding agents use tools and an iterative loop to understand code, plan, and execute tasks, mimicking human engineers.
• Import AI 453: Breaking AI agents; MirrorCode; and ten views on gradual disempowerment — AI systems are demonstrating advanced capabilities in complex coding and face diverse attack vectors as they become autonomous agents.
• OpenAI Has Some Catching Up to Do — Claude Code's "vibe coding" approach is gaining traction over OpenAI's more structured, senior-engineer-focused Codex.
• 👨‍🔧 OpenAI launches Codex app to gain ground in AI coding race — AI coding tools enhance speed but may hinder skill development if over-relied upon for core tasks.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Simon Willison's Weblog.