How do we make an internal AI tool usable for employees with different accessibility needs?
Treat accessibility as a requirement for the workflow, not a feature to add after launch. Before building, identify the essential tasks employees need to complete and the devices and assistive technologies they use. Avoid assuming that everyone can comfortably read a dense chat transcript, use a mouse, hear audio, or work in the same language.
Ask the forward-deployed engineer to use accessible interface patterns and test the actual tool. Depending on the workflow, that can mean keyboard navigation, clear labels for screen readers, readable contrast, captions for audio, and plain-language instructions. If the tool generates content, make it possible to review and edit that content in a familiar format.
Keep a workable alternative available while issues are being fixed. An employee should not have to disclose a diagnosis to a broad group just to complete routine work. Establish a private route for requesting support or reporting a barrier, and make someone responsible for resolving it.
Test with appropriate assistive technology and, where employees choose to participate, with people who use different access methods. Do not claim the tool is compliant simply because an automated checker passes it. The human focus described in the EOS video "Focus on the Human Side of Business" is a useful reminder: an efficient process only works if the people responsible can actually use it.
Category: AI that stays human