DEV Community article highlights accessibility flaws in AI-generated HTML code

DEV Community features a story titled "When coding agents write HTML, what automated validators miss" that looks at accessibility failures in code generated by AI coding assistants. The article explains that when developers ask a coding agent to build a "Save" button, the agent sometimes returns something like <div class="btn" onclick="save()">Save</div>. This markup is, on the surface, perfectly acceptable. Run it through the Nu HTML Checker, and it receives a passing verdict. Run axe-core against it, and the scan reports no violations at all. As a result, developers can end up shipping the code believing two well-known tools have cleared it. In practice, however, the button does not work for many users. Pressing Tab does not move focus to it, and pressing Enter produces no response. keyboard users, along with people relying on screen readers, are therefore unable to operate the Save button at all. The article attributes the problem to the markup choice itself: a div with a click handler is well-formed HTML, which is why Nu passes it, and axe-core finds no violation to flag. Yet neither validator checks the deeper question of whether a keyboard user or screen reader user can actually reach and trigger the element. The piece closes with a practical warning for developers who adopt AI-generated code: a design may look correct because it is syntactically valid and free of automated violations, while failing at the accessibility layer underneath.