Designing for the Next Billion Users in AI
Research
Company
Designing for the Next Billion Users in AI
How we're approaching accessibility and inclusive design as artificial intelligence becomes the primary interface for work.
Our early performance budgets were written for a fast connection on recent hardware. When we instrumented real sessions, the median experience looked nothing like our test environment — and the gap widened precisely in the markets we were trying to reach.
Cut the critical path We deferred everything that was not required to render the first answer, which reduced time-to-first-insight substantially on slow connections. Cache the shape, stream the values Layouts are stable and values are not. Caching structure separately from data let repeat visits render instantly and fill in as data arrived. Degrade, don't block When a model call is slow, the interface now shows the last known result with a clear staleness indicator rather than an empty loading state.
We now report p75 time-to-first-insight by region rather than a single global average. Averages hid exactly the users we most needed to see.
"If your product is only usable on the hardware your team carries, you have built a demo, not a platform."
No spam. Unsubscribe anytime.