The AI assistant that learned to say no

A quiet update changed how millions of people argue with their software — and not everyone is thrilled.

By Dana OyelaranAugust 30, 20265 min read
Robot arm interacting with a glowing touchscreen interface in a lab
Robot arm interacting with a glowing touchscreen interface in a lab

The release notes ran to two sentences. The behaviour change was enormous: the assistant now pushes back, asks for clarification and occasionally refuses outright.

Power users noticed within hours. Support forums filled with screenshots of an assistant politely declining tasks it happily performed a week earlier.

Friction as a feature

The engineering argument is that a confident wrong answer costs more than a clarifying question. The user argument is that nobody asked for a colleague.

Early telemetry reportedly shows fewer corrections and longer sessions, which is exactly the result the team predicted.

Whether that holds once the novelty wears off is the question everyone in the sector is watching.

Enjoyed this story? Get the bolininnan briefing in your inbox.

More to read