Individual vs Organizational AI Gains

Ethan Mollick's productivity paradox: AI demonstrably boosts individual performance and adoption is high, yet organizational performance and macro wage/hours data barely move — because individual gains don't translate to organizational gains by default.

The paradox

Three things are true at once — AI boosts individual work, lots of people use it, and big gains are available — alongside a fourth: companies report only small/moderate gains, with "no major impact on wages or hours worked" as of end-2024 (NBER). The reconciliation: organizational gain requires organizational innovation (rethinking incentives, processes, the nature of work), and that capability atrophied while firms outsourced it to consultants and enterprise-software vendors. Generalized, off-the-shelf playbooks won't work because no one — not even the AI labs — knows your context.

Why it matters

This is the empirical anchor for Leadership, Lab, and Crowd (the prescribed remedy) and a key piece of evidence in AI and Employment — Replacement vs Expansion: macro labor data hasn't shifted yet, which complicates both the "mass replacement" and "explosive expansion" stories — the gains are real but stuck at the individual level.

Independent support (2026)

Anthropic — How Scientists Use Claude reaches the same reconciliation from economic history: "adoption can substantially lag capabilities." Diffusion is slow (Comin & Hobijn 2010), and "productivity gains tend to materialize only after organizations restructure around them" (David 1990; Brynjolfsson, Rock & Syverson's productivity J-curve, 2021). Its scenarios therefore assume wide adoption: even today's AI could roughly double output in CS and statistics, but only if researchers "apply it to every step where it is useful today." At the task level, individual adoption is already rational: use closely follows where AI works (AI Use Tracks AI Capability). The gap opens above the individual.

Connections

• AI-Accelerated Science — science-specific version: scenario gains depend on adoption catching up with capability.
• AI Use Tracks AI Capability — individual task-level adoption follows capability, so the gap isn't individual reluctance.
• Leadership, Lab, and Crowd — Mollick's mechanism for converting individual → organizational gains.
• Secret Cyborgs — hidden individual use is one reason org-level gains stay invisible.
• AI and Employment — Replacement vs Expansion — "no major impact on wages or hours (end-2024)" is a measured data point for that debate.
• Every — a counter-example of an org that did capture gains by reorganizing.

Sources

• Making AI Work — Leadership, Lab, and Crowd — Mollick's four facts and their reconciliation.
• Anthropic — How Scientists Use Claude — "adoption can substantially lag capabilities"; the J-curve citations.