EXECUTIVE PERSPECTIVE · AI & DIGITAL
Closing the AI Productivity Gap
Capabilities are rising fast. Business impact is still muted. The gap is the opportunity—and the risk.
Raw AI capabilities are advancing rapidly on benchmarks and in research labs. Measured economic impact—productivity, employment structure, firm-level returns—remains comparatively muted. That gap is not a reason for complacency. It is the same pattern seen with electricity and other general-purpose technologies: a long period of investment and reorganization before gains show up at scale.
The organizations that close the gap first will capture disproportionate value. Those that treat AI as a series of isolated experiments will not.
Augmentation versus automation
Not every AI application destroys demand for human work. When AI makes a role more productive and demand for the output is elastic, employment in that role can rise. When AI primarily substitutes for tasks and demand is inelastic, employment falls. The practical implication: design AI systems to augment high-value work and expand capacity where the market will absorb more output, not only to cut headcount in exposed tasks.
The J-curve of general-purpose technology
Historically, major technologies required deep complementary investment—new processes, new skills, new organizational designs—before productivity accelerated. Electricity took decades to transform factory layout and output. AI is moving faster, but the complementary work is still required: process redesign, reskilling, product innovation, and measurement systems that capture real value rather than activity.
From tasks to systems
Occupations are bundles of tasks. AI currently handles some tasks well and others poorly. The leadership question is not “Will this job disappear?” but “Which tasks will shift to machines, which will remain human, and how do we redesign the role and the workflow around that split?” The highest-leverage human work increasingly sits at the ends of the process: defining the right question and evaluating whether the answer is actually useful.
Managing fleets of agents
As AI agents improve at execution, more professionals will operate as the “CEO” of a set of agents—scoping problems, directing work, and judging results. That skill is learnable. It combines domain judgment, technical fluency, and the discipline of clear problem definition. Organizations that build this capability systematically will outpace those that only deploy tools.
What leaders should do now
• Connect AI investment to core business problems, not side experiments.
• Measure value in business outcomes, not only model performance.
• Invest in the complementary changes—process, skills, incentives—that turn capability into productivity.
• Design roles around augmentation and judgment, not only automation.
• Treat the next three to five years as the window in which the productivity gap either closes in your favor or widens against you.
Source materials
Informed by research and public commentary from Erik Brynjolfsson (Stanford Digital Economy Lab), including work on AI and employment (“canaries in the coal mine”), the productivity J-curve of general-purpose technologies, augmentation versus automation, and the emerging role of humans in directing and evaluating AI agents.