Complementary Investments

In the AI Story

The four categories operate at different timescales and face different obstacles. Organizational restructuring happens at the speed of internal political negotiation — fast in principle, slow in practice because it disrupts existing power distributions. Human capital development runs at the speed of learning, which for complex skills is measured in years. Business process reengineering runs at the speed of operational change management, which requires both understanding and will. Institutional adaptation — regulations, professional standards, credentialing — operates on the slowest timescale of all, because institutions are designed to resist premature change.

The electric motor example provides the clearest illustration of what happens when complementary investments are skipped. Factories that installed electric motors as direct replacements for steam engines in existing layouts captured modest productivity gains. Factories that redesigned their floor plans around the distributed-power flexibility that individual motors enabled captured transformative gains. The difference was not the technology. It was whether the factory was rebuilt around what the technology made possible.

The AI transition is recapitulating this pattern at compressed speed. Most organizations in early 2026 are layering AI onto existing processes — the equivalent of bolting electric motors onto steam-era shafts. The results are positive but modest. The few organizations undertaking genuine redesign report results that make the incremental adopters' gains look trivial. The gap between these two approaches is the largest source of variance in AI outcomes, and the variance is explained almost entirely by the quality and depth of intangible capital formation.

The practical implication is that support for organizational redesign is not a luxury but the primary determinant of whether the AI transition produces broadly shared gains or concentrated ones. Brynjolfsson has proposed publicly funded organizational innovation centers — modeled on agricultural extension services during the mechanization of farming — to help small and medium-sized firms undertake redesigns they lack the resources or knowledge to implement independently.

Origin

The concept emerged from Brynjolfsson's firm-level empirical research in the 1990s, which required finding variables that explained why identical IT investments produced radically different productivity outcomes across firms. His work with Lorin Hitt at the University of Pennsylvania Wharton School, documented in their 2000 Journal of Economic Perspectives paper Beyond Computation, established the central empirical finding: IT investment produced returns several times larger in firms that had also invested in organizational practices like decentralized decision-making, team-based production, and employee training.

The intellectual lineage extends to economic historians of technology like Paul David and Nathan Rosenberg, who had documented the general pattern that transformative technologies require complementary investments across multiple decades of scholarship. Brynjolfsson's contribution was to provide rigorous firm-level quantitative evidence for the pattern and to integrate it into a predictive framework capable of anticipating the challenges of subsequent technology transitions, including AI.

Key Ideas

Four categories of investment. Organizational, human capital, process, and institutional — each operating at different timescales and facing different obstacles.

Investments are mostly intangible. The capital being built is organizational knowledge, not physical equipment, which makes it invisible to standard accounting.

Technology alone is insufficient. The firms that buy the tool without restructuring the work capture only a fraction of the potential gains.

Variance across firms is enormous. Identical technology produces radically different outcomes based on complementary investment quality.

Policy implication: support redesign. Public investment in organizational innovation is the highest-leverage intervention for widely distributing AI gains.

Debates & Critiques

The debate focuses on whether complementary investments can be accelerated fast enough to match AI's speed of advance. Skeptics argue that organizational learning has irreducible minimum timescales that compressed deployment cycles cannot shorten. Brynjolfsson's response has been to advocate for institutional support — innovation centers, training programs, measurement reform — that can compress the learning curve even if it cannot eliminate it. A separate debate concerns whether the investments are properly characterized as complementary at all, or whether they constitute the actual productive capacity, with the technology playing a merely facilitating role.