How We Actually Build Things
Our process starts with a lot of questions and prototyping. We usually build three or four different approaches to any AI feature before settling on the one that actually works well on mobile devices.
Testing happens continuously—we have a group of volunteers who use early versions of everything we build. Their feedback often completely changes our direction, and we've learned to embrace that rather than fight it.
By late 2025, we're planning to expand our testing program to include more diverse user groups. Real-world usage patterns teach us things that lab testing never could.