A developer's attempt to use Claude Code for migrating jQuery code to vanilla JavaScript ended in frustration. Despite detailed instructions and helper functions in a comprehensive `CLAUDE.md` file, the AI agent made numerous errors across 67 files, demonstrating fundamental misunderstandings of JavaScript basics and DOM manipulation.

Alex Yumashev, a heavy AI user who credits these tools with 10x productivity gains, documented a series of failures. The AI wrote `<script type="module">` tags while immediately using `document.currentScript` (which doesn't work in modules), referenced non-existent DOM elements, and ignored null-checking requirements. It also created CSS selectors with IDs starting with digits (invalid in CSS but valid in jQuery) and failed to handle AJAX-loaded HTML script execution properly.

This real-world test challenges the hype around AI coding agents. While AI excels at greenfield development, it struggles with existing codebases containing legacy patterns and constraints. The experience suggests that for complex refactoring tasks in production applications, human developers remain essential. As Yumashev notes, those maintaining real software for paying customers can likely keep their jobs for another couple of years while AI catches up to the messy realities of brownfield development.