The cleanest story about artificial intelligence is that it makes workers more productive. Higher productivity expands the economy's capacity to produce, lowers unit costs, and creates room for faster growth without equivalent inflation. If AI is a transformative general-purpose technology, that story may eventually be right.
But “eventually” does a lot of work. An economy can spend years investing in a technology before the measured productivity payoff arrives. During that interval, optimism itself can change demand.
The economy can react to future productivity today
Recent Federal Reserve research uses a standard macroeconomic model to examine a “TFP news shock”: households and firms become convinced that productivity will be higher in the future even though current productivity has not yet improved. The result is intuitive but easy to overlook. Expectations of future abundance can create current spending and investment.
Firms build data centers, buy chips, hire specialized workers, and raise capital expenditure. Asset values can rise. Households expecting higher future income may spend more. If productive capacity has not yet caught up, that additional demand can put upward pressure on prices.
The productivity evidence is still developing
The San Francisco Fed has highlighted suggestive evidence that generative AI could function both as a general-purpose technology and as an “invention in the method of invention”—a technology that improves the process of research and innovation itself. If so, the productivity effects could persist rather than simply produce a one-time efficiency gain.
Yet aggregate productivity data do not instantly reveal the impact of a new technology. Adoption takes time. Firms have to reorganize workflows, train employees, redesign products, and learn where the technology is actually useful. The famous productivity paradox around earlier information technology offers a warning against expecting the macro statistics to move at the speed of product demos.
Workers are already responding to the possibility
Boston Fed research based on the New York Fed's Survey of Consumer Expectations found that workers' fears of AI-related job loss increased between December 2024 and December 2025. Those expectations matter economically even before displacement occurs. A household worried about future employment may alter saving, spending, training, or job-search behavior.
At the same time, the September 2026 Beige Book reported broader AI use among firms in the Atlanta Fed district, with contacts deploying AI and automation to improve productivity and efficiency, while most did not expect significant near-term workforce reductions. That combination—rapid adoption without immediate mass displacement—is entirely plausible.
AI complicates the interest-rate story
If AI eventually produces a large positive supply shock, it can reduce inflationary pressure for a given level of demand. But the investment boom required to reach that future can push the other way. Higher expected returns to capital can increase investment demand and potentially put upward pressure on equilibrium interest rates.
There is therefore no simple rule that “AI means lower rates” or “AI means higher rates.” The answer depends on which channel dominates: productivity, investment, wealth effects, labor displacement, fiscal response, or demand for capital.
This is a sequencing problem
The macroeconomics of AI may be less about the final productivity number than about the sequence by which the economy gets there. Imagine a decade in three phases. First comes expectation: firms and investors anticipate transformative productivity. Second comes construction: enormous resources are devoted to compute and complementary infrastructure. Third comes diffusion: AI becomes embedded throughout production and the efficiency gains become visible across industries.
Inflation, wages, employment, and interest rates can behave differently in each phase. A technology can be disinflationary in the long run and inflationary during the buildout. It can eliminate some tasks while increasing demand for other labor. It can raise measured productivity only after years of seemingly excessive investment.
The macro question is bigger than “Will AI take jobs?”
The labor-market question matters, but it is only one component of the adjustment. AI is simultaneously a technology shock, an investment shock, an expectations shock, and potentially a reorganization of how knowledge work is produced.
That is why the most useful macroeconomic analysis of AI should resist single-variable narratives. The same technology can raise expected growth, increase capital spending, alter household behavior, pressure infrastructure, and eventually expand supply. The tension among those forces—not a simple prediction about robots replacing workers—is where the interesting economics begins.