The strongest claims about AI often start with task-level time savings. National productivity is a harder test. It asks whether the economy is producing more output for each unit of labour input, after millions of companies have made their own technology, staffing and investment decisions.
BLS productivity data and BEA output data provide the macro baseline. Neither series can isolate AI on its own, so attribution needs to remain cautious.
Task productivity can appear before national productivity
A worker may complete a narrow task faster while the company spends the saved time on more work, quality control or entirely new activities. Those gains do not always show immediately in measured output.
Implementation costs also arrive first. Businesses may spend on software, training, data and infrastructure before the productivity payoff is visible.
Capital deepening is part of the story
AI is unusually capital intensive at the infrastructure layer. Data centres, accelerators, networking and power investment can raise productive capacity while also increasing depreciation and financing requirements.
The economic question is therefore not simply whether companies buy AI, but whether the resulting output eventually justifies the capital and operating costs.
What would count as convincing evidence
A durable case would combine broader AI adoption, stronger output per hour, investment that translates into revenue or cost savings, and productivity improvements across more than a handful of technology companies.
Until then, national productivity should be treated as an outcome to monitor rather than proof that can be assigned to AI from one quarter's data.