Carson Gross, creator of the htmx library, shares his ambivalent view on AI development tools while walking through a concrete bug fix in hyperscript. The JavaScript-based scripting language suffered a regression in version 0.9.91 where fetch commands incorrectly parsed the 'as JSON' modifier, treating it as a type conversion instead of a response handler.

Using Claude, Gross quickly identified that an overly aggressive refactoring had created a grammar conflict. The shared parse URLOr Expression() method between fetch and go commands now consumed the 'as' keyword as part of general expressions rather than allowing it to function as a fetch modifier. This created a classic parsing ambiguity problem.

When seeking fixes, Claude proposed solutions that were either too hacky or introduced unnecessary complexity. However, Gross knew the parser's existing 'follows' infrastructure—which lets higher-level parse elements claim keywords—could solve this without new state. He implemented a targeted fix that pushed 'as' as a follow token specifically for fetch commands.

The experience illustrates AI's strength in rapid investigation and test generation, but also shows how it falls short on elegant architectural solutions. Developers who don't understand their codebase's existing patterns risk accumulating technical debt when relying on AI for fixes.