What Have We Learned? The Bargain That Built the Middle Class

What Have We Learned? The Bargain That Built the Middle Class

The American middle class was built through a series of negotiated agreements about what employers owed workers, what workers owed firms, and which responsibilities belonged to the government instead of business. Many of those agreements were designed for an industrial economy dominated by large, stable employers. Today, AI, platform work, and the fragmentation of employment are testing whether that model can still hold.

To understand what may come next, it is worth revisiting one of the most influential labor bargains in modern American history and its tradeoffs.

A Settlement, Not a Triumph

By 1950, Walter Reuther had two visions for how to secure American workers. One had just died in Congress and the other got signed in Detroit.

In May 1950, the United Auto Workers and General Motors agreed to a five-year contract. GM workers received a fully funded pension of $125 a month, company-paid health insurance (with GM covering half the premium), cost-of-living adjustments tied to inflation, and an "Annual Improvement Factor" that raised wages four cents an hour each year regardless of bargaining cycles. GM got five years of labor peace and full control over production decisions. Ford and Chrysler matched the model within months. Daniel Bell, then a young Fortune writer, called it the "Treaty of Detroit." The name stuck.

Fortune itself was clear about who got the better deal. "GM may have paid a billion for peace but it got a bargain," the magazine wrote in 1950. The architects of the postwar compact understood at the time that something was being traded for something.

What was being traded matters.

What Reuther Actually Wanted

Reuther's earlier vision was broader. He wanted national health insurance, expanded Social Security, and a federal role in industrial investment decisions. He served on Truman's Commission on the Health Needs of the Nation. He spent the late 1940s pushing for the Wagner-Murray-Dingell bill, which would have established universal health coverage financed through payroll taxes.

That bill died in 1949. Truman's national insurance vision collapsed under organized opposition from the American Medical Association, which ran what was then the largest lobbying campaign in American history. In 1954, the new Internal Revenue Code confirmed that employer-paid health benefits were excluded from employees' taxable income. A dollar of coverage bought through your job was now worth more than a dollar of wages spent on coverage yourself, locking in the logic of getting health insurance from an employer rather than from the government. 

With the federal door closed, Reuther moved to the bargaining table and won at the firm level what he could not win at the national level. Auto workers got the benefits. A national system never came.

These were called "fringe benefits" because they were originally understood as supplements to an anticipated public welfare state. When that broader system never arrived, the fringe became the whole garment.

The World That Followed

The Golden Age is real. Between 1948 and 1973, productivity and the hourly compensation of a typical worker grew in tandem. Real median household income roughly doubled. Union membership reached a postwar peak around 1954 at roughly a third of the private-sector workforce. Defined-benefit pensions, employer-paid health insurance, paid vacation, and seniority-based job security became standard inside large unionized firms, then spread to non-union firms competing for the same workers.

For those who were covered, the compact worked. One income could support a household with a mortgage and a college fund.

The Limits of the Postwar Compact

American manufacturers in the 1950s and 1960s faced almost no global competition, since Europe and Japan were still rebuilding. Profit margins were wide. A few dominant firms could absorb generous contracts without losing market share. Productivity gains funded wages and shareholder returns simultaneously. The compact rested on a market structure that turned out to be temporary. In other words, the postwar bargain was not self-sustaining. It depended on economic conditions that policymakers and executives mistakenly treated as permanent.

It also rested on exclusions the May post traced in detail. Most agricultural and domestic workers sat outside the framework. Black workers were systematically routed into the lowest-paid jobs even inside unionized firms. Women's wages were set as a "secondary" income against a presumed male breadwinner. The Golden Age was broadly distributed within a narrowly defined group.

What Germany Did Instead

In April 1951, while American unions were negotiating firm-by-firm, the West German Bundestag passed the Montan-Mitbestimmungsgesetz. It required coal, iron, and steel companies to seat workers on their supervisory boards at parity with shareholders. In 1976, the Codetermination Act extended the principle to all firms with more than 2,000 employees, with workers holding just under half of supervisory board seats. German workers were given a permanent voice inside the company itself, not a contract that could expire.

The consequences have been long-running. German manufacturing employment held up considerably better than its American counterpart through five decades of globalization. Wages have tracked productivity more closely. Worker training is jointly funded by firms, unions, and the state. Plant closures still happen, but they happen with worker representatives in the room when the decision is made.

Reuther studied the German model and pushed for something similar in the United States. He could not get it and ultimately the American compact was built without it.

Codetermination is not universally portable. It rests on a political coalition and an industrial base the United States never had. Still, the contrast is real. Germany built worker power into how companies are governed; the United States wrote it into contracts that expire. One design has aged far better than the other.

Why the Compact Unraveled

The Treaty of Detroit was made on three assumptions: large firms would stay large, those firms would remain profitable enough to fund their benefit obligations, and the workforce would have long tenures inside them.

By the mid-1970s, all three started falling apart.

Oil shocks, Japanese and German competition, and stagflation compressed margins in exactly the industries that had carried the compact. Deindustrialization accelerated through the 1980s. Steel, autos, and heavy manufacturing shed millions of jobs. There was no federal floor to catch them. Workers laid off from GM did not move into a national health system because there wasn't one. They moved into a private market priced for those still employed by large firms.

The Shift From Shared Risk to Individual Risk

Defined-benefit pensions began converting to defined-contribution plans after ERISA passed in 1974, with the explicit aim of protecting pension promises that companies could no longer reliably keep. The 401(k) provision was added almost by accident in 1978 as a clarification of executive deferred compensation rules. It became the dominant retirement vehicle within a generation. Investment risk shifted from firms to individuals.

Since 1979, productivity has grown roughly 90 percent while typical worker pay has grown about 33 percent by EPI's standard measure.* Economists disagree on the magnitude of that gap, and a fuller note on the methodology appears at the end of this post. No major analysis disputes the direction. The close link between productivity and typical workers' pay that held from 1948 to 1973 has not been restored under any measurement approach.

The compact's design contained its own fragility. Benefits attached to firms expired when firms restructured. There was no public infrastructure to catch what fell off the bargaining table. This is the governance flaw at the center of the American employment model: protections were attached to institutions rather than to people.

The features most often missed about the Golden Age (employer-provided healthcare, pensions, career employment) were tied to a single thread - each one ran through the employer. When the employer changed shape, all of them came loose at once.

What This Has to Do With AI

The compact built for the assembly line was already poorly fitted to a service economy. It is even more poorly fitted to AI-shaped work.

Three pressures collide. Work is fragmenting, with more tasks performed across multiple platforms by people who have no single employer. Firms are using AI to compress middle-skill cognitive work the way earlier automation compressed middle-skill manual work. And the early gains from AI are accruing disproportionately to capital rather than labor. Productivity data is showing acceleration in some sectors whereas wage data is not.

The Same Bargain, Smaller Firms

The most concrete contemporary version of Reuther's bargain is happening in Hollywood. In 2023, the Writers Guild of America struck for 148 days and SAG-AFTRA for 118 days, with generative AI as a central issue in both. Both unions ratified contracts addressing it directly. AI cannot be credited as a writer. AI-generated material does not count as "literary material" that displaces writer compensation. Actors must give informed consent for "digital replicas" of their voice or likeness, with separate compensation for each use. Studios cannot scan a background actor for half a day's pay and then use that replica without limit (Perkins Coie analysis). These are real protections, won at the bargaining table when federal policy offered none.

They are also industry-specific protections, won by two unions covering roughly 175,000 members in one sector. They do not cover the freelance writers, voice actors, animators, illustrators, or translators outside the guilds doing parallel work for different companies. They do not address what AI developers did with the training data already scraped. They expire in three years. If the structure of the entertainment industry reorganizes around AI workflows, as it very likely will, the protections may not survive the reorganization.

The UAW's 2023 Stand-Up Strike produced a parallel result in another industry: a 25% wage increase, restoration of cost-of-living adjustments, an end to wage tiers, and commitments to bring battery plants under the national agreement as the industry electrifies. The contracts expire on April 30, 2028, a date set deliberately to enable coordinated action across the Big Three. Whether the Big Three still hold their current share of a partially electrified, increasingly AI-managed auto market by then is something the contract cannot guarantee.

These are Reuther moves, and good ones. They are also Reuther moves in the architectural sense: firm-by-firm, industry-by-industry, with no broader floor underneath them.

The US and Europe Are Diverging Again

International policy is again taking a different shape. The EU's Platform Work Directive and Spain's Ley Rider create employment protections that attach to the worker rather than to a single firm. The EU AI Act takes a different route, classifying AI systems used in hiring, management, and firing as "high-risk" and requiring transparency and human oversight wherever they are deployed. The OECD's 2024 work on AI in the workplace treats algorithmic transparency, worker consultation, and human-in-the-loop requirements as baseline expectations. At the US federal level, the October 2023 executive order on AI (which included worker-impact provisions) was rescinded in January 2025, leaving most of this to firms and to state action.

If the lesson of 1950 is that firm-level bargaining can win real gains in a narrow window, the lesson of 1979 onward is that firm-level gains do not survive industry restructuring. The compromise is harder still under current conditions, because there is no "GM of AI" on which to anchor an industry-wide compact. Hollywood writers are not auto workers. Software engineers are not steel workers. The firms restructuring around AI are smaller, more fragmented, and faster-moving than the Big Three were in 1950. The bargaining unit that could hold them accountable, in most cases, does not exist yet.

Reuther's question (what should be settled inside a single firm versus what should be settled across the economy) is back. The answer in 1950 was "as much as we can win at the table." The answer for 2026 is yet to be written.

Disclosure: Stela is a UAW member through her adjunct faculty local, which informs how she thinks about which workers keep their protections when an industry restructures, and which workers lose them.

Your Sphere of Influence

The Perkins move (April post) was setting standards before legislation made them required. The Reuther move is similar in form and different in scope: when the broader system will not deliver, build what you can inside your reach and design it to outlast you. Then keep pushing for the broader system anyway.

What Reuther Did	What You Might Try
Negotiated firm-level benefits when national policy stalled, with the explicit understanding that the firm-level deal was second-best	Map which parts of your employment offer (health, retirement, training, leave) depend entirely on your company continuing to exist as it currently does. Where could a portable or pooled structure protect workers if your firm restructures or is acquired?
Tied wage growth to productivity through an explicit formula (the Annual Improvement Factor), making the link contractual rather than discretionary	Audit how your compensation system actually shares productivity gains. Is the link mechanical, discretionary, or absent? Communicate the answer to employees.
Pushed for worker representation in production decisions, lost on that point in 1950, kept pushing for it for the next two decades	Identify one operational or strategic decision (AI deployment, workforce planning, scheduling) where worker representatives could sit in the room when the decision is made. Try it. Document what changes.
Linked the UAW's bargaining to broader social goals (civil rights, healthcare, training) so the union's gains were not isolated from public policy	Pick two or three public policy questions in your sector (AI governance, benefits portability, skills standards). Decide where your firm's interest aligns with a broader policy fix, and back it.
Treated the Treaty of Detroit as a temporary settlement, not a finished compact, and reopened terms as conditions changed	Build a rhythm of revisiting your employment compact every two or three years rather than letting it ossify between crises.

What This Is Really About

The compact that built the American middle class was a particular instrument, designed by particular people for the conditions of a particular moment. When those conditions changed, the compact dissolved, because there was no underlying public infrastructure to fall back on.

The question for HR leaders is whether the structures being built today rest on conditions that will hold for thirty years. If they rest on the firm staying intact, the industry staying stable, or the workforce staying long-tenured, that is Reuther's compromise without Reuther's political project to back it up.

Reuther sat across from three companies and knew what he was getting and giving up. The compact forming around AI has no such table, no such counterpart, and no such clarity about the terms.

---

This is a post in our 2026 series "What Have We Learned?" Subscribe to receive monthly explorations of how America has defined what employers and employees owe each other, and what you can do about it.

*A note on the productivity-pay gap

The 90 percent productivity, 33 percent compensation figure cited above comes from EPI's standard analysis, which measures net productivity at output prices against the compensation of production and non-supervisory workers deflated by the CPI-U-RS (Bivens and Mishel 2015; EPI 2024 update). Other economists have argued this overstates the gap. The main critiques: (1) productivity and compensation should be deflated using the same price index, since using output prices for productivity and consumer prices for wages creates a wedge from price-index differences alone (Lawrence 2016; Strain 2019, AEI); (2) total compensation including benefits has grown faster than cash wages, so wage-only measures understate worker gains; (3) average compensation should be compared to average productivity (or median to median) for consistency.

Anna Stansbury and Lawrence Summers, in their 2017 NBER paper, examined the question using three-year moving averages and found that "over 1973-2016, one percentage point higher productivity growth has been associated with 0.7 to 1 percentage points higher median and average compensation growth." Productivity growth still matters substantially for typical workers. They also estimated that if productivity growth from 1973 to 2016 had matched the earlier postwar period, median and mean compensation would have been roughly 41 percent higher in 2016, holding inequality constant. The largest single contributor to whatever gap remains, in their analysis and in EPI's, is rising compensation inequality. The top earners' pay has tracked productivity but the median worker's has not.