600 Visual Accessibility Reports Mined From Copilot, Cursor, Claude Code, Codex, and OpenCode Issue Trackers
The first empirical study of visual accessibility in AI developer tools screened 2,652 keyword-retrieved candidates from issues and forum discussions across five ecosystems — GitHub Copilot in VS Code, Cursor, Claude Code, OpenAI Codex, and OpenCode — and used a three-model ensemble to identify 600 unanimously positive reports. Barriers cluster into screen-reader incompatibility, visual presentation problems like contrast and differentiation, and readability limits specific to AI interfaces such as chat panels, terminal agents, generated diffs, and streaming status output. The authors argue the record reflects both interaction design and the reporting and maintenance culture of each surrounding ecosystem, which is a pointed finding given how fast these surfaces ship.
↳ Follow the thread