"Productivity isn’t everything, but in the long run it is almost everything."
That famous observation by economist Paul Krugman captures the central preoccupation of macroeconomics. Economists measure the contribution of technological innovation to economic growth through "total factor productivity" (TFP), which tracks the sustained ability of an economy to produce more output using the exact same amount of inputs.
As such, TFP is arguably the most critical measurement in the economist’s toolkit. The continual, compounding process of producing more with less is the primary mechanism through which human living standards improve over time. A country’s capacity to elevate its standard of living depends almost entirely on its ability to steadily raise its output per worker. Technology is the catalyst that makes that happen, and total factor productivity is the metric used to track it.
To grasp the tangible economic weight of technology-generated productivity, consider a recent research paper published by the National Bureau of Economic Research (NBER). The study estimates that an additional 0.5% of annual TFP growth would be enough to completely stabilize the United States government’s finances, locking in the debt-to-GDP ratio right at today’s levels.
That figure—just 0.5%—might not sound like much at first glance. Yet, if that modest acceleration could be sustained over the course of the next decade, the NBER estimates it would reduce the baseline forecast for US government debt by a staggering $2 trillion.
Looking further out over a 30-year horizon, a sustained 0.5% boost to TFP would leave the US government’s debt-to-GDP ratio 42 percentage points lower than the NBER’s baseline forecast. Compared to the organization’s pessimistic forecast, that same productivity boost would make the debt-to-GDP ratio 80 percentage points lower.
Given the seemingly hopeless state of modern government finances and the relentless expansion of national deficits, maintaining today’s level of indebtedness looks like a dream scenario—the kind of outcome that feels far too good to be true.
However, researchers at AI safety and development firm Anthropic believe the economy might be on the precipice of achieving something even greater.
Anthropic recently conducted a comprehensive study analyzing 100,000 real-world conversations on Claude.ai. The objective was to estimate how long the various tasks embedded in these conversations would take to complete both with and without AI assistance, and to study the subsequent productivity implications across the broader economy.
The conclusion drawn from the research was striking: large language models (LLMs) could potentially raise total factor productivity by a full 1.1 percentage points.
If an annual increase of 0.5% is enough to stabilize US government finances for decades, the implications of a 1.1% boost are profound. Such an acceleration would likely alter the fiscal landscape entirely, resolving budgetary pressures that currently defy conventional solutions.
Naturally, there are compelling reasons to approach such optimistic forecasts with a healthy dose of skepticism.
The Anthropic study demonstrates, for instance, that Claude can save educators an average of four hours of labor by generating comprehensive lesson plans and curricula in a mere 11 minutes. Yet, translating those kinds of isolated time-savings into broader, macroeconomic output requires economic modeling laden with best-guess assumptions and the risk of false precision.
Even if Anthropic’s calculations regarding time-savings are entirely accurate, the company’s model could still misjudge the ultimate productivity outcome. It is entirely possible that society will choose to reallocate the time saved by artificial intelligence into pursuits that are economically unproductive, such as watching more short-form video content or consuming additional media.

In that scenario, the integration of AI would successfully elevate human welfare by granting individuals more free time, but it would fail to expand aggregate wealth through increased economic output. That outcome would still represent wonderful news for individuals seeking a better work-life balance, but it would offer little solace to governments desperately hoping for a silver bullet solution to their structural debt crises.
Conversely, there are equally strong arguments suggesting that Anthropic’s economic model may be far too pessimistic.
As the researchers note in their methodology, the study intentionally avoids factoring in the rate of technological adoption or the much larger productivity multipliers that will inevitably arise from the deployment of vastly more capable future AI systems.
In other words, the study’s baseline assumption is that humanity continues to utilize artificial intelligence exactly as it does today, relying on today’s language models without further improvement for an entire decade.
Because language models have demonstrated a track record of growing noticeably better every few months, and because the global workforce has only just begun to understand how to effectively harness them, Anthropic’s researchers acknowledge that their estimate may represent an approximate lower bound on the true productivity effects of AI.
If a 1.1% increase truly represents the floor for AI-induced productivity gains, society might simultaneously pay down long-standing government debt while carving out ample leisure time. And that projection accounts solely for the impact of AI on non-physical, digital work—a dynamic that will expand exponentially once advanced robotics enter the commercial market.
To dismiss this wave of technological optimism entirely is to bet that the trillions of dollars corporations currently plan to pour into artificial intelligence capital expenditures and research and development will be entirely wasted.
That outcome remains within the realm of possibility, as historical technological revolutions have rarely unfolded on a neat, predictable schedule.
Yet the most compelling reason for optimism is that Anthropic’s 1.1% estimate is derived strictly from AI making existing tasks faster to complete, without accounting for the technology’s capacity to entirely transform how those tasks are structured and executed.
Historically, transformative productivity improvements—stemming from monumental shifts like electrification, enterprise computing, or the birth of the internet—did not stem merely from accelerating old tasks. Instead, they drove economic miracles by fundamentally reorganizing production.
While it is impossible to accurately model these nascent, reimagined workflows in advance, the historical precedent suggests that the aggregate impact will easily surpass what researchers have attempted to measure so far.
Anthropic’s researchers are deliberately careful to caveat their hopeful findings, taking pains to enumerate the limitations of their methodology and document the myriad assumptions underpinning their models. Furthermore, even if all those assumptions prove correct and AI-driven productivity solves the US government’s structural debt problem, political realities suggest lawmakers might simply spend their way right back into fiscal jeopardy.
Even so, given the near-universal pessimism surrounding the trajectory of global fiscal health, even a small probability that Anthropic’s projections prove accurate warrants a reassessment of conventional economic assumptions. The financial position of the United States government may not be as intractable as contemporary consensus suggests, and the long-term outlook for the US dollar may not be as perilous as many fear.
In the long run, productivity remains the ultimate driver of human progress—and artificial intelligence may be on the verge of making the global economy significantly more productive.

Comments (0)