Beyond benchmarks
Model benchmarks tell you how smart a system is. They say nothing about whether it makes work feel better. So alongside the usual numbers, we run what we call calm sessions.
What a calm session looks like
Volunteers spend an hour on real work with Cirro switched on, then an hour with it switched off. Afterwards we ask simple questions about focus, stress and interruptions.
A release only ships if people report fewer interruptions with Cirro than without it. That rule has saved us from more than one clever feature.
