Inclusivity by Design
If it doesn't work for everyone, it doesn't work. Practical inclusivity for agentic products.
Inclusivity is a quality metric
Inclusive design isn't a separate workstream you do after the feature ships. It is one of the definitions of "works."
A model that performs brilliantly for one demographic and poorly for another isn't "mostly working." It is broken for the people it doesn't serve, and you should treat it that way.
A practical playbook
- Disaggregate your evals. Don't just report aggregate accuracy. Slice by language, region, dialect, ability, age, device.
- Include disabled users in research. Not as a checkbox at the end, as participants from week one.
- Test with assistive tech. Screen readers, voice control, switch input, high contrast. If your AI UI breaks under a screen reader, the whole feature is broken.
- Write copy for the widest plausible reader. Plain language, defined acronyms, no idioms that don't travel.
- Localize, don't just translate. A literal translation of an English UI is rarely an inclusive UI.
Where AI specifically helps
Agentic AI is genuinely amazing for accessibility, voice interfaces, summarization, real-time translation, alt-text generation. We have a responsibility to ship those wins, not just talk about them.
Build for the edges and the middle takes care of itself. Build only for the middle and you leave people behind.