AI products engagement metrics fail due to friction
The author argues that traditional software engagement metrics like session duration and prompt volume fail for AI products, where high usage can signal frustration and friction rather than value. The piece highlights how teams are shifting to measure actual progress and outcomes instead of raw activity, pointing to examples like Lovable's agent 'vent tool' and Cursor's analytics. The author concludes that the next chapter of software design is about making tools less necessary to use through better human-AI interaction models. Open article →