Usage vs Utility
One of the most persistent errors in technology analysis is the conflation of usage with utility. The entire apparatus of technology measurement — daily active users, monthly active users, session lengths, engagement rates — rests on the implicit assumption that usage is a reliable proxy for value. The assumption held through the first decade of the commercial internet, when people went online to accomplish specific tasks. The assumption broke during the social media era, when platforms achieved unprecedented usage numbers alongside declining well-being among heavy users. The AI transition is producing a subtler and potentially more consequential divergence. Where social media's usage-utility gap was primarily a gap of time allocation — opportunity cost — AI's potential gap is a gap of capability development. When a person uses AI to produce work they could not produce alone, their output improves while their opportunity to develop the underlying capability diminishes. The data captures the first. It cannot capture the second.
In the AI Story
The distinction matters because the mechanisms of divergence are different. Social media's usage-utility gap was produced by engagement optimization — platforms optimized for time-on-platform regardless of whether the time produced value users themselves recognized. Users remained the judges of value; they simply weren't equipped to resist the manipulation.
AI's potential gap operates through a different mechanism: capability substitution. The tool does not consume time that could have been spent otherwise. It substitutes for cognitive development that would have occurred through struggle. The productivity appears. The underlying capability does not.
The parallel to prior automation dependence patterns is instructive. When GPS navigation became ubiquitous, drivers stopped developing spatial navigation skills — and studies documented measurable declines in wayfinding ability among younger users who had never needed to navigate without the tool. AI operates across a wider range of cognitive tasks than GPS, producing a broader and deeper potential for capability atrophy.
Meeker's quantitative framework is structurally vulnerable to this kind of invisible cost. The framework sees the usage. It sees the productivity gains. It does not see the capability that was not developed, the expertise that was not built, the judgment that was not cultivated, because these absences do not generate data points.
Origin
The concept emerged from Meeker's long tracking of the divergence between social media engagement metrics and well-being metrics across her Internet Trends reports. The 2025 AI report extends the framework to a phenomenon whose divergence operates through a different mechanism.
The distinction gains particular urgency in AI analysis because the stakes extend beyond time allocation to the formation of human capability itself — the capacity to do work that, over time, produces expertise rather than merely output.
Key Ideas
Usage measures frequency; utility measures value. The two correlate in some contexts and diverge in others, and technology analysis that conflates them produces misleading conclusions.
Social media revealed the divergence. The platform that maximizes engagement is not the platform that maximizes user well-being; engagement optimization and utility production are different objectives.
AI's gap operates through capability substitution. Where social media consumed time, AI may consume the opportunity to develop capability — a more consequential loss because it propagates through all future work.
The data cannot see absence. Quantitative metrics capture what happens; they cannot capture the capability that was not developed, the understanding that was not built.
The distinction reshapes policy questions. If usage reliably indicated utility, growth data would settle debates about AI's value. Because it does not, the debates must continue on other grounds.
Appears in the Orange Pill Cycle
Further reading
- Mary Meeker, Trends — Artificial Intelligence (Bond Capital, 2025)
- Jonathan Haidt, The Anxious Generation (Penguin Press, 2024)
- Cal Newport, Deep Work (Grand Central, 2016)