AI Code Assistant Fails to Remove jQuery Without Breaking Core Functionality

In a revealing test of modern AI code assistants, software developer Alex Bilbie of JitBit attempted to automate the removal of jQuery from a legacy web application using Claude Code—an AI tool designed to assist with code refactoring. The outcome, as documented in Bilbie’s blog post, was far from seamless: the AI failed to properly replace key jQuery functions, introduced syntax errors, and omitted critical event handlers, rendering parts of the application non-functional. The experiment underscores a growing tension between the promise of AI-driven development and the nuanced reality of maintaining legacy JavaScript codebases.

Bilbie’s project involved a moderately complex frontend built with jQuery 1.11, featuring DOM manipulation, AJAX calls, and event delegation across multiple modules. He instructed Claude Code to "remove all jQuery dependencies and replace them with native JavaScript equivalents." The AI responded by converting simple selectors like $("#button") to document.getElementById("button"), which appeared correct on the surface. However, it missed the broader context: jQuery’s cross-browser normalization of event properties, asynchronous handling of DOM-ready states, and the use of .on() for delegated events on dynamically loaded elements. In several instances, the AI replaced .click() with .addEventListener('click'), but neglected to account for event bubbling or the absence of jQuery’s event object normalization, leading to uncaught exceptions in older browsers.

Further complications arose when the AI attempted to refactor AJAX calls. While it successfully converted $.ajax() to fetch(), it omitted error handling blocks, failed to convert JSON parsing logic, and ignored the need for polyfills in environments without native Promise support. One particularly egregious error occurred when the AI removed a jQuery UI widget initialization—replacing it with an empty function—thereby breaking the entire UI interaction flow without warning. The tool also did not identify or remove jQuery’s script tag from the HTML header, leaving a dead dependency in place.

The incident has sparked debate among developers on Hacker News, where the original blog post garnered over 50 points and 64 comments. Many respondents noted that while AI tools have made impressive strides in generating boilerplate code and suggesting optimizations, they still lack deep contextual understanding of legacy systems. "AI doesn’t understand why jQuery was used in the first place," wrote one user. "It sees syntax, not intent."

Experts in software modernization argue that the failure reflects a broader industry challenge: automated refactoring tools often operate on syntactic patterns without grasping architectural intent. "jQuery wasn’t just a library—it was a bridge to browser compatibility in the early 2010s," explained Dr. Lena Torres, a software archaeologist at MIT. "Removing it isn’t a mechanical substitution. It’s a cultural and technical migration that requires understanding of historical constraints, team knowledge, and user expectations."

Bilbie concluded his experiment by manually refactoring the codebase over two days, documenting each change and testing across three browser environments. He noted that while AI can accelerate simple tasks, complex legacy migrations still demand human oversight. "The AI didn’t fail because it was dumb," he wrote. "It failed because it was asked to do something that requires wisdom, not just pattern matching."

This case serves as a cautionary tale for organizations considering AI-assisted legacy modernization. While tools like Claude Code, GitHub Copilot, and Amazon CodeWhisperer show promise in greenfield development, their application in high-stakes, legacy-heavy environments must be approached with rigorous validation, human-in-the-loop review, and clear scope boundaries. As AI becomes more embedded in software workflows, the line between assistant and architect remains firmly in human hands.