Erik Brynjolfsson

In the [YOU] on AI Field Guide

The Orange Pill opens with individuals experiencing gains so dramatic they seem to defy the aggregate statistics—a single engineer producing what would previously have required a team, a non-technical founder prototyping a product over a weekend, a backend developer building interfaces for the first time. Brynjolfsson’s framework explains the discrepancy without dismissing either experience. The individuals have already made the complementary investments—the process redesign, the skill development, the workflow restructuring—that the aggregate economy has not yet undertaken. They are living on the far side of the J-curve, in the zone where the technology’s potential is being realized, while the average organization remains in the dip. The paradox has returned, wearing new clothes.

His February 2026 claim in the Financial Times that U.S. productivity had jumped roughly 2.7 percent in 2025—nearly double the preceding decade’s annual average—was the most contested empirical assertion of the early AI era. Apollo Chief Economist Torsten Slok quipped that “AI is everywhere except in the incoming macroeconomic data,” consciously echoing Solow’s original formulation. Brynjolfsson’s response was structural: the harvest phase had begun for organizations that had invested in redesign; the dip persisted for those that had merely purchased the tool. The gap between the two populations was widening with each generation of AI capability.

The cycle reads Brynjolfsson as the economist who supplies the distributional question that pure productivity analysis omits. The great decoupling—productivity rising while median wages stagnate—has characterized the American economy since the late 1970s. The AI transition, if its gains follow the same distributional lines as previous digital technology gains, will accelerate the erosion. Brynjolfsson’s prescription is specific: redesign the incentive structure, reform the measurement system, invest in education at a scale matching the magnitude of the change, and choose augmentation over automation wherever the choice can be made. Technology is not destiny.

Origin

Brynjolfsson arrived at MIT’s Sloan School in the early 1990s with deep econometric training and a genuine fascination with information systems. He understood that information technology behaved differently from previous capital: it scaled differently, it interacted with organizational structures in ways that standard production functions could not capture, and its value depended almost entirely on the quality of the complementary investments that surrounded it. The gap between the visible proliferation of computers and the invisible absence of productivity gains struck him as a measurement and organizational problem, not as evidence that the technology was unproductive.

His resolution—the timing problem, the measurement problem, and the organizational problem—had a prediction built into it that proved accurate across every subsequent technology wave. The electric motor had followed the same J-curve, spending decades bolted onto steam-era factory layouts before the factories were rebuilt around the motor’s logic. The personal computer followed the same pattern. Each time the aggregate statistics disappointed before the complementary investments matured and the gains arrived. By the time large language models began transforming knowledge work, Brynjolfsson had a framework calibrated across more than a century of technology transitions.

His landmark 2023 study, “Generative AI at Work,” co-authored with Danielle Li and Lindsey Raymond, provided the first rigorous micro-level evidence about how AI interacted with worker skill levels. Studying 5,179 customer service agents, they found a 14 percent average productivity gain—but the distribution was radically uneven: novice workers improved by 34 percent, experienced workers saw minimal impact. The finding suggested something hopeful and something troubling simultaneously, a tension Brynjolfsson has been navigating in public ever since.

Key Ideas

The Productivity J-Curve. Transformative technologies trace a specific arc through productivity statistics: an initial dip during which investment costs are real but returns are invisible, followed by gains that eventually surpass the pre-technology trend. The dip is not evidence of failure but of investment in intangible capital—organizational knowledge, new skills, redesigned processes—that does not appear in standard accounts. The J-curve is not a counsel of patience. It is a diagnostic tool with an urgent prescription: invest in the complementary assets now, because the length of the dip depends on the speed and quality of those investments.

The Turing Trap. The Turing Test set the wrong goal for AI development: building machines that substitute for human performance rather than machines that amplify it. The Turing Trap is not a conspiracy but an incentive structure—a tax code that subsidizes capital over labor, a research culture that celebrates human-versus-machine benchmarks, an organizational instinct toward automation as the path of least resistance. Escaping the trap requires deliberate design choices at every level: researchers, managers, and policymakers choosing augmentation over displacement wherever the choice can be made.

Intangible capital and the measurement crisis. The most important assets in the modern economy—organizational knowledge, human skill, institutional trust—do not appear on any balance sheet. AI is simultaneously a form of intangible capital and a technology whose value depends almost entirely on the quality of the complementary intangible capital surrounding it. The standard productivity metrics are blind to this dynamic, producing systematic underinvestment in the assets that would unlock the technology’s potential.

Bounty and spread. Every major technology transition creates bounty—the expansion of total economic value—and spread—the distribution of who captures the gains. The two concepts are not in zero-sum contest but they are not automatically aligned. The great decoupling of productivity from median wages since the late 1970s is the historical evidence that bounty does not automatically translate into broad prosperity. Achieving broad prosperity has always required deliberate institutional construction: education systems, labor protections, and social insurance built through political struggle.

The race between education and technology. Drawing on Claudia Goldin and Lawrence Katz, Brynjolfsson argues that the distribution of AI’s gains will be determined by a single variable: whether education produces workers with the capabilities the AI economy rewards faster than the technology commoditizes those capabilities. The skills AI substitutes are precisely the ones the current educational system trains. The skills AI complements—judgment, integration, the capacity to ask the right question—are the ones the system is least effective at developing.