AI futures: Planning for transformative scenarios before they hit

What if the next five years reshape the global economy more than the last fifty? The answer depends on whom you ask.

Traditional economists and policymakers in Washington see artificial intelligence (AI) as a general-purpose technology akin to electricity or the internet, with benefits unfolding gradually. Productivity gains, in this view, depend on complementary investments in skills, infrastructure, and institutions. History is the guide: Past innovations took decades to diffuse and deliver broad-based growth. Let us call this the Washington Consensus on AI.

The San Francisco Consensus on AI sees it differently: Scaled-up models and data and improving algorithms will soon deliver transformative AI, possibly even superintelligence, capable of remaking economies and societies on a much shorter timeline. In this view, breakthroughs will arrive soon, rapidly and recursively, making the coming years potentially among the most consequential in centuries. This is not science fiction. AI systems have already become remarkably capable at programming—for example, Anthropic's Claude Code, an AI coding assistant, was used to build Claude Cowork, a desktop automation tool, demonstrating that AI can now substantially contribute to the creation of new AI products. Such developments raise the real possibility of recursive self-improvement, in which AI systems accelerate their own advancement and significantly increase growth.

Given these divergent perspectives, it is prudent to engage in scenario planning and hedge across multiple plausible futures rather than betting on a single one. We recently helped convene a workshop on the global economic and financial implications of AI, organized by the Economics of Transformative AI (EconTAI), a research initiative at the University of Virginia, and the International Monetary Fund (IMF). This laid the groundwork for a careful scenario planning exercise, organized together with the Windfall Trust, that asked participants to ponder how the rapid technological advances that are foreseen in the San Francisco Consensus would transform our economy. Why focus there? If the Washington Consensus on AI is right, existing policy frameworks can adapt incrementally as they have before. The real preparedness gap lies in the alternative: rapid transformation that outpaces the resilience of current institutions.

Two divergent futures

The workshop explored scenarios assuming that AI systems and robotics could be capable of performing most economically valuable tasks by 2030—a timeline chosen not as a prediction but as a stress test. Some technologists would consider this pace modest; many economists consider it unusually fast. Within this technological baseline, the key uncertainty becomes how societies, firms, workers, and governments respond. A critical question in this context is the speed of AI adoption and diffusion. 

Our workshop explored two scenarios for 2030, both assuming equally advanced AI capabilities:

• The first scenario assumes uneven adoption. Large firms and tech hubs integrate AI, but smaller businesses, governments, and developing economies lag behind. Regulatory friction, public skepticism, and skill shortages slow diffusion. Productivity improves, labor market disruption occurs but is manageable, and fiscal and monetary frameworks remain largely intact.
• A runaway scenario supposes that AI adoption accelerates at breakneck speed, penetrating nearly every sector of the economy. Automation becomes ubiquitous. Jobs are displaced faster than labor markets can adapt. Economic power concentrates in a few firms and locations. Inequality spikes. Fiscal systems strain under eroding labor tax bases and rising social demands. Central banks face new challenges as inflation dynamics shift, and financial markets grow volatile.

Themes and transition costs

The main objective of the scenario planning exercise conducted by the workshop was to wrestle with the questions that will arise in the described futures, not to achieve a consensus on how to respond. AI could substantially boost aggregate output in both scenarios—but the gains may be concentrated by country or region, and may leave emerging markets and low-income countries further behind.

Fiscal systems may need to adapt. In the United States, for example, approximately three-quarters of federal tax revenue comes from labor income. As labor's share of output declines, traditional tax bases may erode. Moreover, many countries may struggle to tax AI-driven gains given mobile capital and the concentration of AI profits in a small number of jurisdictions. Meanwhile, demands for expenditures may rise—to support displaced workers, build skills, and strengthen social insurance systems. Detaching safety nets from employment—providing basic benefits regardless of job status—may become necessary. Such a shift would reduce work incentives—an undesirable outcome today when economies depend on human workers, but potentially necessary in a future where AI and robots perform most economically valuable tasks (Era Dabla-Norris, Daniel Garcia-Macia, and Anh D. M. Nguyen, forthcoming).

Monetary policy may also face new challenges. Rapid productivity growth could push up the neutral interest rate—the rate that maintains full capacity utilization and stable inflation—altering the calibration of policy. AI-driven technological advances may give rise to stark relative price changes, including changes in the relative value of labor, complicating monetary policy decisions, even though efficiency gains may exert disinflationary pressure. Traditional measures of labor market slack, such as the Phillips curve—tracking the typically inverse relationship between inflation and unemployment—may become less relevant for inflation dynamics, requiring central banks to develop new indicators of economic capacity utilization.

Financial markets may become more volatile as heavy investment in AI infrastructure requires large amounts of financing and may create new pockets of vulnerability due to risky financing arrangements such as high levels of debt—highly levered systems are vulnerable to sudden shifts in expectations, even when long-run fundamentals remain solid. Concentration among a small number of firms may raise systemic risks. The transition period may pose acute challenges—including stranded assets, debt overhang, and the potential for sharp corrections if expectations disappoint. 

A recurring theme was that the adjustment period is likely to involve significant disruption. The displacement of traditional workers and firms could outpace the adaptation of our institutions and generate social and political pressures that complicate policy responses. Managing this transition will be critical to ensuring that the advances in our technological capabilities also advance human well-being.

The value of preparation

Scenario planning is not about predicting the future—it's about equipping institutions and policymakers with frameworks and tools that help them respond across a range of plausible outcomes. What indicators should we track as the AI transformation accelerates? Are social protection systems ready for a possible surge in unemployment? Can central banks manage the relative price pressures that may occur? Are regulators prepared for AI-driven systemic risks?

These are no longer abstract questions—they are stress tests for the resilience of our economic frameworks. As AI capabilities evolve, this work must continue. Iterative scenario planning, institutional stress-testing, and adaptive policy frameworks will be critical to navigating the AI transition.

Whether AI adoption and diffusion is uneven or pervasive, as laid out under the two scenarios above, the macroeconomic consequences will be profound. Given the uncertainties involved in predicting the precise path of technology, the challenge is to prepare for all of them and steer the transition toward broad, shared prosperity.

Policy priorities and indicators to watch

As a practical takeaway, policymakers may want to begin developing playbooks for the following challenges. This involves:

• On the monetary side, exploring new frameworks for assessing aggregate demand and inflation dynamics when traditional labor market indicators diverge from AI-driven productivity trends.
• On the fiscal side, reexamining social insurance models and considering tax systems or other capital-sharing mechanisms that can capture a broader share of AI-driven value creation without introducing significant distortions.
• On the regulatory side, recognizing that AI's scale and data advantages may reinforce winner-take-most dynamics, with implications for prices, inequality, and the tax base, and considering how existing competition and data-governance frameworks might adapt to these shifts.
• On the financial side, strengthening supervisory capacity to monitor leverage, concentration risks, and new forms of interconnectedness arising from AI-related investments.

Preparing for the future also requires close attention to real-time signals. To gauge whether we are heading toward the Washington or San Francisco consensus on AI, policymakers and observers could track a range of indicators: (1) AI research breakthroughs, particularly advances that automate the AI research process itself, signaling recursive self-improvement; (2) AI product releases and their performance across cognitive, reasoning, and robotics benchmarks; (3) AI diffusion and deployment across the economy, monitoring which sectors are being transformed, what segments of the labor market are impacted and how quickly; and (4) financial market trends, such as investment flows into AI, data center build outs, and the stock valuations of AI-related companies, which reflect collective expectations about future AI capabilities. The pace of change across these indicators could help signal when to shift from incremental adaptation to more transformative policy responses.

Era Dabla-Norris is deputy director at the fiscal affairs department of the International Monetary Fund.