Image Not FoundImage Not Found

  • Home
  • Work & Leadership
  • Fed Chair Kevin Warsh Gives Employers a Practical AI Test: Productivity Boom, Job Shock, or Investment Letdown?
Two office workers look at a laptop screen on a modern office floor, with colleagues at desks behind them.

Fed Chair Kevin Warsh Gives Employers a Practical AI Test: Productivity Boom, Job Shock, or Investment Letdown?

Federal Reserve Chairman Kevin Warsh used opening remarks at a new Productivity and Jobs Task Force conference on November 16 to do something business leaders have been waiting for from policymakers: treat AI not just as a future productivity lift, but as a live labor-market and capital-allocation problem.

Warsh sketched three possible paths for what the federal government currently calls “super intelligence,” or SI. In the best case, the technology raises productivity, growth stays above trend, and the gains are shared between workers and owners of capital while the economy remains near full employment. In a second path, GDP still rises, but workers are displaced faster than companies and labor markets can absorb them, pushing unemployment above levels consistent with full employment. In the third, heavy investment and swelling equity valuations are followed by disappointing returns, falling stock prices, slower growth, and higher unemployment, with the later recovery failing to deliver the promised productivity payoff.

That matters because companies are making hiring, redesign, and spending decisions now, before the macroeconomic evidence is settled. The useful question is no longer whether AI is impressive. It is how to tell whether an organization is turning model capability into durable output, whether workers are moving into better roles or out of the workforce, and whether investment is being supported by real demand rather than hope.

A Fed framework for the workplace, not a forecast

Warsh was careful about what he was not doing. He did not announce a rate move, change Fed policy, assign probabilities to the three paths, or offer a national estimate for job loss or job creation. He described the scenarios as illustrative starting points. He also said the conference was one of five task forces he has convened as chairman to broaden the Fed’s field of vision beyond daily market data.

Still, the speech is significant because the chair explicitly linked AI adoption to employment and financial markets at the same time. Same-day reports from AP and Axios highlighted the same point: Warsh discussed the possibility of widespread job loss, a stock-market decline, and higher unemployment if AI investment returns disappoint.

He also made a more ambitious economic argument. SI, he said, may differ from earlier general-purpose technologies because it could accelerate the creation of ideas, not just automate existing tasks. That goes to a long-running problem in economics: research effort has increased substantially over time even as research productivity has fallen. If AI changes the speed and method of innovation, the upside is not limited to trimming labor costs. It could affect total-factor productivity, living standards, and the long-run growth rate.

But the speech also drew a firm line between machine capability and human responsibility. Warsh said human judgment, morality, and accountability remain distinct. For employers, that is not a philosophical aside. It is a management constraint. Companies still have to decide where automation is appropriate, where human review stays essential, and how work gets reorganized around those boundaries.

What companies should measure before cutting or hiring

The most practical reading of Warsh’s remarks is as a scorecard for management.

First, measure output per worker or revenue per worker, but pair it with customer outcomes. A team that produces more drafts, tickets, code, or reports is not necessarily more productive if accuracy, conversion, retention, or service quality deteriorates. Demo counts and pilot announcements are weak evidence. Repeatable output that customers will pay for is stronger evidence.

Second, track worker transitions, not just headcount. Are employees whose tasks are increasingly automated being redeployed, trained, and promoted into higher-value work? Or are they being released because the organization changed the workflow faster than it created new roles? The same underlying AI capability can produce very different labor outcomes depending on those choices.

Third, watch wages and hiring by task. If AI is complementing labor, companies should start to show demand for people who can supervise systems, interpret results, manage exceptions, and handle customer situations where judgment matters. If it is mostly displacing labor, hiring may narrow while wage pressure weakens in tasks the technology can perform more cheaply or more quickly.

This is the central tradeoff for leadership teams. A company that cuts roles before it has proved workflow gains may boost near-term margins while losing institutional knowledge and weakening future demand if displaced workers pull back spending. A company that treats every existing task as untouchable may preserve stability while missing real efficiency gains. The durable test is whether leaders can redesign jobs and controls so AI complements human judgment and creates a credible path into new work.

Investors need a utilization story, not just a valuation story

Warsh’s third scenario is especially important for boards, lenders, and investors because it shifts attention from AI enthusiasm to AI economics.

Large data-center and model investments depend on assumptions about future utilization, pricing power, and customer demand. If those assumptions hold, capital spending can support a broad productivity story. If they fail, the damage does not stop at a lower tech multiple. It can spill into private credit, hiring plans, supplier orders, and business investment more broadly.

That means investors should watch whether AI-related revenue is keeping pace with capital spending, whether computing capacity is being productively used, and whether financing is being extended on the basis of recurring demand or projected demand. A rising stock price alone does not prove AI is creating lasting value. A falling stock price alone does not prove the technology has failed. The key question is whether cash-generating use cases are materializing fast enough to justify the buildout.

The broader Fed context reinforces the point without settling it. AP noted that Warsh did not address inflation in this speech, even though other officials have discussed AI-related demand for chips, electricity, copper, and other inputs. Vice Chair Philip Jefferson said in a November 13 speech that workers specializing in tasks AI can perform more efficiently may face displacement, and that a lasting productivity increase could raise firms’ demand for funding and put upward pressure on interest rates. Those are his views, not a binding FOMC position, but they show how many channels are now in play at once.

What Warsh did not do is just as important as what he did. He did not identify which industries, occupations, or firms are most exposed. He did not say when measured productivity would reflect any gains, whether GDP statistics would capture quality improvements quickly, or how the Fed would translate these scenarios into forecasts or rate decisions.

That leaves companies, workers, and investors with a harder but more useful assignment. Do not ask whether AI is real. Ask whether it is producing broad output gains, whether workers are moving into better-paid and more durable roles, and whether the revenue base is strong enough to support the capital being committed. If enough firms get those decisions right, Warsh’s optimistic scenario becomes more plausible. If they do not, a productivity story can turn into a labor shock, a capital-markets shock, or both.