Ami Brabson's Work in HCI: What It Actually Means for the Field
Ami Brabson has spent the better part of two decades working on practical problems at the intersection of computing and human behavior. She was at Microsoft for a long time, then moved to Google where she continued pushing accessibility and wearable interaction forward. Her contributions aren't the kind of thing that makes headlines, but they're the kind of thing that makes tools actually usable for people who aren't typing on a keyboard. Most people in HCI know her work through accessibility research, specifically around mobile and wearable interfaces for users with visual and motor impairments. That's the surface level. The deeper impact is in how she approached problem-space definition — she didn't just build features for disabled users, she redesigned interaction models so that accessibility became a first-class constraint rather than an afterthought. That shift changed how teams at both companies think about prototyping. I remember working on a voice-guided navigation feature for a mobility-impaired demographic, and the initial design kept failing because nobody had considered that screen readers and voice assistants compete for the same auditory output channel. Ami's earlier papers on this exact problem helped me realize we needed a haptic feedback fallback layer before we even started coding. It saved us about three weeks of dead-end development. Without that reference point, we would have shipped something that worked for sighted users and broke for everyone else.
One thing most beginners miss about her approach is that she treats context switching as a primary friction point. When designing for users who rely on assistive technology, every additional mode a person has to mentally switch between compounds the cognitive load exponentially. Her wearable computing research from the mid-2010s showed that reducing modal switches by even one per task could improve task completion rates by 40% for certain populations. That's not a small number. Most teams I talk to still design for single-mode interaction and add multimodal features as an enhancement rather than a necessity. Her Google years focused heavily on on-device machine learning for interaction prediction. The idea is straightforward: instead of sending user behavior data to the cloud and waiting for a model to process it, you run lightweight models directly on the device. This cuts latency from around 300 milliseconds to under 50, which is the difference between an interface feeling responsive and feeling broken. The trade-off is that model accuracy drops slightly without cloud-based refinement, so she advocated for a hybrid approach where frequently used patterns are cached locally and rarely seen behaviors get synced back for retraining. It's not elegant, but it works in production. There are real limitations to this framework though. On-device ML requires significantly more hardware budget, and not every product can afford to optimize for it. Teams with tight memory constraints or lower-end device targets will find that the accuracy trade-off becomes unacceptable. In those cases, a server-side model with aggressive caching and prefetching can get you 70% of the latency benefit at 20% of the cost. It's worth knowing both paths before committing.
Another area where her influence is measurable but hard to quantify is in how accessibility testing has moved from a compliance checkpoint to a design input. Before her work gained traction, most teams ran accessibility audits at the end of a sprint. Now it's common to see assistive technology users involved in early concept testing at Google and Microsoft. The metric that changed was bug discovery timing — issues that used to show up during QA now surface during prototyping. That means fewer reworks and tighter shipping cycles. If you want to dig into the actual research, most of her peer-reviewed papers are available through IEEE Xplore and ACM Digital Library. The Google Research blog also published several accessible summaries of her wearable interaction work around 2018 to 2020. There isn't a single canonical resource, which is probably why the knowledge hasn't spread further into the general HCI community. That's something the field still needs to fix. The practical takeaway is this: if you're building any interface that involves mobile or wearable input, start by mapping out every context switch a user with assistive technology would need to make. Then remove at least one. Ami's work gives you a framework for doing that without reinventing the wheel. It won't make your product accessible to everyone, but it will stop you from making it inaccessible to the people who need it most.
Get the Full Details
