The Wage Slowdown Is Rewriting the Labor-AI Narrative

The Wage Slowdown Is Rewriting the Labor-AI Narrative

The latest signal from the wage data is easy to understate because it arrives as a deceleration rather than a break. Wage growth for blue-collar workers, service workers, sales workers, administrative workers, and support workers has slowed toward roughly 3%. More importantly, it now appears to trail wage growth for management and professional workers. That ranking challenges two popular stories that shaped the post-pandemic labor debate. The first story said that lower-wage, nonprofessional occupations had gained a more durable bargaining position after the pandemic. The second said that those same occupations were relatively insulated from artificial intelligence because AI would mainly pressure knowledge work.

Both stories now look too simple. The post-pandemic wage surge at the lower end of the labor market increasingly looks less like a permanent structural repricing of work and more like the result of an unusually tight labor market. When vacancies were abundant, quits were high, stimulus had repaired household balance sheets, and employers had to staff reopened services, workers with modest wages had credible outside options. Pay rose because the market was short labor. Once the labor market cooled, the bargaining premium faded. That does not mean the earlier wage gains were fake. It means the mechanism was cyclical tightness, not necessarily a lasting shift in the institutional power of lower-wage labor.

The second story is being challenged from the other side. AI and automation are not stopping at professional writing, coding, research, and analytics. They are spreading into routine office work, call centers, sales support, scheduling, claims processing, bookkeeping, warehouse management, manufacturing quality control, logistics routing, customer service, and other areas that sit between manual labor and white-collar cognition. The old distinction between “knowledge workers exposed to AI” and “frontline workers protected from AI” is becoming less useful. Many lower- and middle-wage roles contain repeatable information tasks, coordination tasks, monitoring tasks, or customer-interface tasks that can be partially automated or tightly supervised by software.

If this trend persists, it changes the macro interpretation of labor-market cooling. Slower wage growth in these occupations may moderate wage-driven inflation. It may also support productivity growth if firms use automation to produce more output with fewer hours or with slower wage pressure. But it weakens the idea that deglobalization, reduced immigration, and AI insulation would automatically create a durable lower-wage labor renaissance. The more accurate thesis is conditional: workers can win when labor markets are tight, but technology, automation, and employer bargaining power can still cap the persistence of those gains when the cycle turns.

The Two Narratives That Are Being Tested

The first narrative was born out of the remarkable labor market of 2021 and 2022. Lower-wage occupations saw unusually strong wage growth because firms could not hire fast enough. Restaurants, hotels, warehouses, retailers, trucking firms, local services, and health-support businesses all faced staffing pressure. Workers switched jobs aggressively. Employers that once assumed abundant labor had to raise wages, improve schedules, offer signing bonuses, or tolerate higher churn. For a while, it looked as if the bottom of the wage distribution had finally regained leverage.

That interpretation had a real empirical basis. Wage growth was stronger for job switchers than for job stayers. Lower-wage workers gained more than higher-wage workers in several measures. The Beveridge curve shifted in ways that suggested firms were posting far more vacancies per unemployed worker than in the pre-pandemic expansion. In plain language, labor scarcity forced firms to bid for workers who had previously been treated as easy to replace. The result was one of the more meaningful short-run compressions of wage inequality in recent U.S. history.

But a wage surge caused by tightness is not the same as a permanent regime change. A permanent regime change would require institutions and market structure to change in ways that keep workers’ outside options strong even after demand normalizes. That could come from unions, sectoral bargaining, durable labor shortages, licensing barriers, immigration constraints, skill scarcity, or legally stronger worker voice. A cyclical wage surge only requires more vacancies than available workers. Once vacancies fall, quits slow, hiring standards rise, and employers regain patience, wage growth can decelerate quickly.

The second narrative was that AI would mainly hit professional occupations. The early public demonstrations of generative AI made this intuitive. The technology could draft memos, summarize documents, write code, analyze text, generate images, and answer complex questions. That looked like a direct challenge to lawyers, consultants, analysts, software engineers, designers, marketers, and administrators. By contrast, workers who cooked food, stocked shelves, delivered goods, repaired equipment, cleaned buildings, cared for patients, or operated machinery seemed physically shielded. A model can write an email, but it cannot unclog a pipe.

The problem is that many lower-wage jobs are not purely physical. A warehouse role includes scanning, routing, picking optimization, inventory checks, and performance monitoring. A retail role includes scheduling, checkout, customer triage, and replenishment. A customer-service role is often a structured conversation around a known menu of problems. An administrative-support role contains document intake, data entry, calendar logic, invoice reconciliation, and compliance routing. Even manufacturing increasingly includes machine vision, predictive maintenance, robotics coordination, and software-driven quality control. AI does not need to replace an entire job to weaken wage growth. It only needs to automate enough tasks, increase monitoring enough, or expand the pool of workers who can do the remaining tasks.

Why the Post-Pandemic Wage Surge Was Mostly a Tightness Event

The strongest evidence for a tightness interpretation is the timing. Lower-wage wage acceleration coincided with a labor market in which employers had to compete aggressively for workers. Demand had rebounded faster than labor supply. Some workers were constrained by health risks, childcare disruptions, immigration interruptions, early retirements, geographic mismatch, and reassessment of service-sector work. Fiscal transfers and stronger household balance sheets gave workers more ability to reject unattractive jobs. Firms that needed immediate staffing had to pay more.

This is classic search-and-matching economics. In the Diamond-Mortensen-Pissarides framework, wages are not set by a frictionless spot market. They are negotiated inside matches. The worker’s bargaining position improves when job-finding prospects are strong and the employer’s cost of vacancy is high. A tight labor market raises the value of the worker’s outside option. It also raises the cost to the firm of leaving a role unfilled. The wage bargain shifts toward labor even if the worker’s fundamental productivity has not suddenly jumped.

A simple numerical example clarifies the point. Suppose a service worker produces $25 of value per hour and previously earned $15 because similar workers were easy to find. If the firm suddenly cannot fill shifts, the worker may earn $18 or $20 because the cost of vacancy is high. That wage increase is real, and it can be socially important. But if the labor market later loosens and the firm can hire again at $16 or $17, the earlier wage level will not necessarily persist. The wage was supported by scarcity, not by an institutional claim on productivity.

The post-pandemic economy also included demand distortions. Goods demand surged, then services reopened. Logistics networks were stressed. Inventories were rebuilt. Firms hired defensively because they feared being short labor. That produced a form of labor hoarding and overstaffing in some areas. As demand normalized and interest rates rose, employers gained time. They could slow hiring without immediate revenue loss. They could reduce overtime, allow attrition, and become more selective. That is exactly the environment in which lower-wage wage growth should decelerate if it was mainly driven by tightness.

This does not mean immigration or labor supply are irrelevant. Reduced immigration can tighten specific labor markets, especially agriculture, construction, hospitality, caregiving, food services, and local services. But immigration is only one part of labor supply, and wage outcomes depend on demand, substitution, productivity, and enforcement. If firms can substitute technology, change schedules, redesign roles, or accept slightly lower service quality, a reduced labor pool does not mechanically create accelerating wages. Labor scarcity must be binding and persistent to keep wage growth elevated.

AI Exposure Is Broader Than the White-Collar Story

The deeper surprise is that AI exposure is spreading into the middle of the occupational map. The standard AI conversation started with college-educated knowledge workers because text-generating systems visibly perform tasks that used to require formal education. But from an economic perspective, the key unit is not the job title. It is the task. Any task with structured inputs, repeatable decision rules, measurable output, and digital records is a candidate for automation or augmentation. Many nonprofessional jobs contain exactly those tasks.

Customer service is the clearest example. A call-center agent appears to be a service worker, but the job is heavily information-based. The worker identifies the customer, classifies the problem, searches a knowledge base, follows a compliance script, updates records, and escalates edge cases. AI can handle part of the conversation directly, suggest responses in real time, summarize calls, score sentiment, and reduce after-call documentation. Even when the human remains, the firm may need fewer agents, less training time, and fewer supervisors. Wage pressure falls because the worker is less scarce.

Administrative and support roles face a similar channel. Scheduling, email triage, document processing, procurement forms, invoice matching, payroll questions, travel booking, compliance reminders, and customer records can be automated or semi-automated. These roles were once protected by organizational complexity: someone had to know where the forms were, whom to email, and how to move information through the firm. AI lowers the value of that procedural memory. It turns internal bureaucracy into searchable workflow.

Sales and retail work are also changing. AI can score leads, recommend next-best actions, generate outreach copy, answer product questions, forecast demand, optimize pricing, and route customers through self-service channels. In retail, checkout automation, inventory systems, labor scheduling, loss-prevention analytics, and customer-service bots reduce the number of hours needed per dollar of sales. The person on the floor still matters for trust, exception handling, and physical execution, but software increasingly determines the workflow.

Manufacturing and warehouse roles illustrate the physical-digital blend. Robotics, machine vision, automated guided vehicles, predictive maintenance, digital twins, and AI scheduling do not remove the need for human labor overnight. They change the number and type of workers required. A warehouse may still hire pickers, but algorithms can set routes, monitor productivity, reduce training time, and increase throughput per worker. A factory may still need operators, but machine vision can catch defects and software can guide preventive repairs. The bargaining value of routine repetition declines when the routine is encoded into systems.

Task Substitution Can Cap Wages Without Causing Mass Layoffs

One mistake in the AI debate is to look only for dramatic layoffs. Technology often affects wages before it affects employment. If automation improves output per worker, firms may slow hiring, reduce overtime, consolidate support roles, or avoid replacing people who leave. Employment can look stable while the wage path weakens. This is especially plausible in service and support occupations where turnover is high and firms can adjust through attrition.

The marginal worker matters. Wages are set by the scarcity of the next available worker, not only by the productivity of existing workers. If AI allows a firm to train workers faster, use less experienced workers, offshore some support, or centralize tasks in shared-service centers, the effective supply of labor rises. A larger effective labor pool reduces wage pressure even if headcount does not collapse. That is why AI can be disinflationary through labor-market channels without producing an immediate unemployment shock.

Consider a customer-support team of 100 agents. If AI tools let each agent handle 15% more tickets, management may not fire 15 people tomorrow. But it may stop hiring, allow attrition to take headcount down to 90, reduce overtime, and pressure future wage increases. The official employment data may show a gradual slowdown rather than a sudden rupture. The wage data will often detect the pressure earlier because the firm no longer needs to bid as aggressively.

This is consistent with the concept of routine-biased technological change. Earlier waves of software and automation did not only replace factory workers. They reduced demand for routine clerical and production tasks while increasing demand for abstract problem-solving and some nonroutine personal services. AI extends that logic because it can process language, images, and decisions that were previously hard to automate. The boundary between routine manual and routine cognitive work is moving.

The effect also interacts with monitoring. Digital tools allow employers to measure output, pace, error rates, customer sentiment, and compliance in real time. Monitoring can raise productivity, but it can also weaken workers’ discretion. A worker who once had tacit control over workflow may become an operator inside a system designed elsewhere. The productivity gain is captured by the firm if the worker’s unique contribution becomes less visible or less scarce.

Immigration, Labor Supply, and the Substitution Margin

Reduced immigration was expected by some observers to support wages for lower-wage occupations because immigrants are disproportionately represented in several labor-intensive sectors. The logic is straightforward: if labor supply falls, wages should rise. But the actual wage outcome depends on elasticity. Firms can respond to labor scarcity by raising wages, raising prices, reducing output, changing quality, automating tasks, reorganizing work, or lobbying for different labor channels. The more substitution options firms have, the weaker the wage response.

In sectors with truly binding physical labor needs, immigration restrictions can still raise wages or create shortages. Construction, caregiving, agriculture, food preparation, maintenance, and local services cannot be fully virtualized. But even there, firms can substitute in partial ways: modular construction, scheduling software, self-service ordering, robotic cleaning, warehouse automation, prepackaged food, remote monitoring, or greater use of capital equipment. The point is not that technology eliminates labor. The point is that technology changes the wage elasticity of labor scarcity.

The post-pandemic period may have temporarily hidden that substitution margin because demand was so strong and firms had little time to redesign processes. They needed workers immediately. Over time, high wages and high turnover became incentives to invest in systems. A restaurant that cannot find workers experiments with kiosks. A retailer adds self-checkout and inventory automation. A logistics firm uses route optimization. A back office adopts AI document processing. What began as a labor shortage becomes a capital-deepening cycle.

This is why the current wage deceleration matters. If lower immigration alone were enough to create persistent wage acceleration in these occupations, wage growth should remain clearly elevated. Instead, slowing toward roughly 3% suggests that the labor-market tightness component has faded and the substitution component is becoming more important. Reduced labor supply may still matter, but it is not dominating the broader forces of cooling demand and automation.

For inflation, this is important. A labor-scarcity regime with limited substitution is inflationary because firms pay more and raise prices. A labor-scarcity regime with rapid substitution can be less inflationary because firms invest in productivity and reduce labor intensity. The current data lean toward the second interpretation, though the transition can be uneven across sectors.

Wage Inflation, Productivity, and Unit Labor Costs

The macro channel from wages to inflation runs through unit labor costs. A firm’s labor cost per unit of output depends on wages, hours, and productivity. If wages rise 5% and productivity rises 1%, unit labor costs rise quickly. If wages rise 3% and productivity rises 2%, unit labor costs are much calmer. Slower wage growth in lower-wage occupations therefore matters because these occupations are heavily represented in labor-intensive services, where inflation persistence has been a concern.

A simple formula is useful: unit labor cost growth is approximately wage growth minus productivity growth. If wage growth for a broad group of occupations slows toward 3% while productivity growth improves because of automation, the inflation impulse from labor costs can decline sharply. This does not mean inflation disappears. Rents, energy, medical costs, tariffs, supply shocks, margins, and taxes can still matter. But the labor-cost channel becomes less threatening.

The Federal Reserve cares about this because wage growth is both a price-pressure variable and a demand variable. Slower wage growth reduces the risk of a wage-price spiral. It may give policymakers more confidence that inflation can settle without a severe unemployment rise. But it also means household income growth may cool, especially for workers with high marginal propensities to consume. The same data point can be positive for inflation and less positive for broad consumption.

This duality is central for markets. Equity investors may welcome slower wage growth because it protects margins and lowers rate pressure. Bond investors may welcome it because it reduces inflation risk. But consumer-facing businesses may see weaker demand if lower- and middle-income wage growth slows. The macro signal is therefore not simply bullish or bearish. It is a reallocation signal: from labor income toward margins and productivity, from wage inflation toward disinflation, and from broad consumption toward more segmented demand.

The productivity side matters just as much. If automation raises output per hour, slower wage growth does not necessarily imply weaker real output. The economy can grow with less labor-cost pressure. That is the attractive soft-landing version of the story. The risk is that productivity gains accrue narrowly while wage income weakens broadly. Then the economy looks efficient on paper but fragile in distribution.

The Distributional Problem Behind a Productivity Upside

The optimistic interpretation is that broader AI and automation adoption will raise productivity, moderate inflation, and support real incomes over time. That outcome is possible. If firms become more efficient and competitive markets pass gains to consumers, real purchasing power can improve even if nominal wage growth slows. If workers use AI tools to become more valuable, wages can recover through complementarity. If new tasks and industries emerge, employment can shift rather than shrink.

But the distributional history of technology urges caution. Productivity gains do not automatically become wage gains. They flow through ownership, bargaining, competition, and policy. If the tools are owned by firms, if the data are controlled by platforms, if workers have weak bargaining power, and if product markets are concentrated, the first-round benefit often appears in margins. Workers may receive some gains, consumers may receive some lower prices, but capital owners capture a large share.

The classic production function language helps. Output depends on labor, capital, technology, and organization. When technology improves, the marginal product of some workers rises, but the marginal product of substitutable tasks can fall. If AI is labor-augmenting, it raises the value of the worker. If AI is labor-replacing, it raises the value of the capital owner. The same aggregate productivity gain can therefore raise wages for engineers, technicians, managers, and entrepreneurs while suppressing wages for clerks, agents, support workers, and routine coordinators.

This is why management and professional wage growth outpacing lower-wage categories is not merely a curiosity. It suggests that scarcity may remain concentrated in judgment-intensive, managerial, technical, and relationship-heavy roles. Workers who can direct systems, interpret ambiguous information, own accountability, or integrate across domains remain valuable. Workers whose tasks are structured and measurable face more substitution. The labor market may not be splitting between physical and cognitive work. It may be splitting between scarce judgment and scalable routine.

For policy, this creates a difficult tradeoff. Slowing wage-driven inflation is welcome, but a regime in which productivity gains bypass broad labor income can increase inequality and political stress. The answer cannot be to reject productivity. The answer is to broaden capture through competition, training, mobility, worker voice, and diffusion of technology beyond a few platforms.

What Investors Should Watch Next

The first indicator is occupational wage dispersion. If the thesis is right, wage growth should remain firmer in roles that combine judgment, accountability, technical skill, and client trust, while routine support roles cool. The relevant comparison is not only college versus noncollege. It is routine versus nonroutine, replaceable versus accountable, monitored versus autonomous, and system-user versus system-designer.

The second indicator is hiring rather than layoffs. AI adoption may first show up as fewer job postings, slower replacement hiring, reduced overtime, smaller entry-level cohorts, and tighter promotion ladders. A company can become more labor-efficient without a dramatic layoff announcement. Investors should watch vacancies, job openings, temporary-help employment, hours worked, and quits by occupation.

The third indicator is margin quality. If slower wage growth is supporting margins because productivity is rising, the market may reward firms with credible automation strategies. But analysts should distinguish durable productivity from simple labor squeeze. A firm that raises margins by improving service quality, lowering errors, and expanding capacity has a better long-term story than a firm that merely underinvests in labor until service deteriorates.

The fourth indicator is diffusion. If small and medium-sized firms gain access to AI tools, competition can increase and productivity gains can spread. If the benefits accrue mainly to large platforms with data, compute, and distribution, concentration may rise. Diffusion determines whether AI becomes a broad economic capability or another moat for incumbents.

The fifth indicator is consumption segmentation. Slower lower-wage income growth should show up in more pressure among households with high debt service, high rent burdens, and low excess savings. That does not necessarily mean aggregate consumption collapses, because higher-income households and asset owners can keep spending. But the composition of demand becomes more uneven. Retail, credit, travel, restaurants, autos, and discretionary services may diverge by customer base.

The sixth indicator is policy response. If AI-enabled automation spreads while lower-wage wage growth cools, expect more attention to wage transparency, noncompete limits, labor classification, immigration design, retraining, data rights, platform power, and antitrust. The productivity debate will become a capture debate. Markets usually notice policy risk late, but the labor data can give an early signal.

The Scenario Map for the Next Twelve Months

The most useful way to treat the wage slowdown is as a scenario variable rather than a point forecast. In the benign scenario, lower-wage wage growth cools because the panic phase of hiring is over, but employment remains firm and productivity improves. Firms use AI and automation to reduce bottlenecks, workers move into higher-value tasks, and consumers benefit from lower service inflation. In that world, slower wage growth is not a recession signal. It is part of a normalization process that allows real rates to fall over time without forcing a collapse in demand.

In the margin-led scenario, the same wage slowdown supports earnings more than households. Firms invest enough in automation to keep output growing, but they do not share much of the gain with labor. Unit labor costs improve, operating margins expand, and equity markets reward companies with credible labor-saving technology. Consumption does not collapse because high-income households and asset owners keep spending, but demand becomes narrower. This scenario can look excellent for indexes while feeling mediocre for the median worker.

In the adverse scenario, wage deceleration is not only automation-driven; it is also a sign that labor demand is weakening. Hiring slows, hours fall, credit stress rises, and lower-income households lose momentum. AI adoption may still raise efficiency in some companies, but the macro effect is overwhelmed by weaker income growth. In that world, the wage slowdown would be an early warning that the economy is moving from disinflation toward demand fragility.

The distinction between these scenarios requires watching hours, vacancies, claims, delinquencies, small-business hiring plans, and sector-level margins together. Wage growth near 3% is not enough by itself to diagnose the regime. If productivity accelerates and employment holds, the signal is constructive. If hours and job openings deteriorate, the same wage print becomes more troubling. The market should resist the temptation to treat lower wage growth as automatically bullish.

There is also an asset-allocation distinction. The benign scenario favors duration, quality growth, and selected consumer exposure because inflation pressure fades without deep recession. The margin-led scenario favors firms that own automation tools, platforms, workflow software, industrial technology, and scaled operating systems, but it increases political and demand-distribution risk. The adverse scenario favors defensives, balance-sheet quality, and a more cautious view of lower-income consumption. The wage data are therefore not merely labor statistics. They are a bridge between inflation, productivity, earnings, and credit risk.

Conclusion: The Labor Market Is Cooling Into a New Technology Regime

The deceleration of wage growth among blue-collar, service, sales, administrative, and support workers is not just a soft labor-market print. It is a test of the story investors and policymakers have been telling about the post-pandemic economy. The evidence increasingly suggests that the earlier wage surge was driven mainly by extreme labor-market tightness. Once that tightness faded, the durable bargaining power of lower-wage work looked less secure.

At the same time, AI and automation are proving broader than the early white-collar exposure narrative implied. They are moving into routine office work, customer service, sales support, manufacturing workflows, and warehouse operations. They do not need to eliminate jobs outright to affect wages. By raising output per worker, reducing training requirements, improving monitoring, and expanding substitution options, they can cool wage growth before employment cracks.

The macro result could be constructive: lower wage-driven inflation, better productivity, and a less painful path for interest rates. But the distributional result is more ambiguous. If productivity gains mainly flow to firms and capital owners, the economy can become more efficient while broad labor income becomes less dynamic. The next regime will be defined not only by whether AI raises productivity, but by whether workers, consumers, firms, and governments share the dividend in a stable way. Wage growth near 3% in the occupations once expected to benefit most from scarcity is an early warning that the capture question is already here.