Web Accessibility in the age of Agentic Coding

Short summary

The 2026 WebAIM Million report shows a 10.1% YoY increase in accessibility errors, reversing years of slow progress — likely linked to agentic coding tools reproducing inaccessible patterns from training data. LLMs over-compensate with ARIA labels, actually increasing errors (59.1 vs 42 per page on non-ARIA sites). The author advocates a shift-left approach using Playwright and axe-core to catch violations during development rather than post-deployment.

• •WebAIM 2026 report shows 10.1% YoY increase in WCAG errors, reversing prior progress
• •AI agents mimic inaccessible code patterns; ARIA overuse worsens the problem
• •Shift-left A11Y testing with Playwright + axe-core catches issues during development

Generated with AI, which can make mistakes.