Redistributed

Article: Doug O'Laughlin. "Engels' Pause and the Permanent Underclass." Fabricated Knowledge (Substack), April 16, 2026. Link: Fabricated Knowledge

Why This Piece Matters

Economic historian Robert C. Allen coined "Engels' Pause" to describe the period from 1780 to 1840 when British working-class wages grew just 12% while per-capita GDP surged 46% — with virtually all gains captured by capital. Doug O'Laughlin, co-founder and president of semiconductor research firm SemiAnalysis, argues that the AI revolution is about to trigger the same dynamic in the information economy. The piece is significant not because the argument is novel — Acemoglu, Korinek, and others have modeled AI-driven labor displacement rigorously — but because it translates academic displacement theory into investor-facing language read by over 87,000 subscribers in the technology and finance sectors. When the people allocating capital start describing mass displacement as a "fiduciary duty," the redistribution question moves from theory to implementation.

The piece also matters for what it reveals about framing. O'Laughlin writes with genuine analytical clarity about who gets hurt, then concludes that "capital could have an incredible decade." That tension — diagnosing the wound while celebrating the weapon — is itself a data point about how the AI transition is being understood by the people with the most power to shape it.

The Argument

The Historical Parallel

O'Laughlin builds his case on Allen's research into the first Industrial Revolution. Between 1780 and 1840, per-capita GDP in Britain grew roughly 46% (Harley and Crafts's estimates), while Charles Feinstein found that working-class wages rose only 12%. The crucial insight: it was not unskilled laborers who suffered most but skilled artisans — handloom weavers who commanded wage premiums of up to 100% over the broader workforce. That premium created the economic incentive to automate them away. Handloom weavers' weekly wages fell from 240 old pence in 1806 to 99 pence by 1820 — halved in a single generation.

The returns during this period flowed almost exclusively to capital. The top 1%'s share of national product rose from 25% in 1801 to 35% by 1848. The economy tripled in output over 80 years, but wages didn't begin rising until after the full displacement of handiwork — and only when an entirely new paradigm, the railroad buildout, created fresh demand for labor.

The Modern Handloom Weavers

O'Laughlin identifies today's equivalent: roughly 70.7 million American workers in "management, professional, and related occupations" — 43.9% of the US workforce. With broader definitions including office, admin, and sales roles, that figure reaches approximately 100 million of the 161.3 million employed. This is the richest segment of the economy, and O'Laughlin argues that its "golden age" ran from 1993 to 2020.

Drawing on BLS occupational statistics and Anthropic's Economic Index, he ranks occupations by displacement incentive — the product of AI task coverage and median wage. Software developers top the list: 94% theoretical AI task coverage at a $133K median wage yields a $125K displacement incentive per worker, totaling $208 billion in compensation at risk across 1.65 million workers. Financial analysts follow at $96K per worker, management analysts at $92K. Across the 18 occupations charted, 16.7 million workers carry $1.03 trillion in annual compensation at risk — 10.9% of the entire US workforce.

"Replacing the $120k a year analyst with $10K in tokens is going to be almost a fiduciary duty," he writes, "and will happen as soon as feasibly possible."

The Trigger: Mythos and Superhuman Information Processing

O'Laughlin frames the Mythos model as the catalyst — a system capable of finding "thousands of critical vulnerabilities that survived decades of human review." He describes this as proof of "human cybersecurity researcher displacement" and argues the John Henry moment has already passed. The implication extends beyond security: legal drafts costing $5,000 in junior associate time will cost $50 in minutes; financial models consuming an analyst's entire day will be generated for pennies.

Projecting the Pause

Extrapolating the Engels' Pause pattern to 2040, O'Laughlin projects: with 2.7% annual productivity growth and just 0.8% compensation growth, GDP per worker reaches roughly $268K per year while worker compensation sits at approximately $87K — a gap of around $28K per worker, roughly $4.6 trillion per year flowing to capital economy-wide. That represents a 53% gap between productivity and wages, dwarfing even the Great Decoupling that began in 1973.

He notes the early signals are already visible — productivity ticking up, layoff announcements soaring, entry-level wage growth stalling — but cautions this is "the smoke and the fire is yet to come."

The Counter-Case: Electrification

O'Laughlin considers whether AI might follow the pattern of electrification rather than the first Industrial Revolution. During electrification, Paul David documented a productivity paradox: electric motors took 30 years to appear in productivity statistics because they augmented an already-adapted industrial labor force rather than replacing it. New tasks — repairing electrical infrastructure — absorbed displaced workers quickly.

But O'Laughlin dismisses the analogy. "When I ask Claude to make a DCF end to end, there is no human in the loop. Versus before I had a human use a human oriented tool (excel) to create and run the analysis." He argues that pre-AI software was closer to a handloom — a human-oriented tool — while AI is closer to the power loom: a wholesale replacement of the process.

What the Piece Gets Right

The artisan parallel is precise. The insight that high-wage skilled workers face the greatest displacement pressure — because the premium itself creates the incentive to automate — is well-grounded in Allen's research and maps cleanly onto the current moment. Junior lawyers, financial analysts, and software developers are today's handloom weavers: skilled, well-compensated, and doing exactly the kind of structured information work that AI handles well.

The displacement incentive framework is useful. Ranking occupations by the product of AI task coverage and median wage produces a concrete, falsifiable prediction about where displacement pressure will hit first and hardest. This is more analytically productive than vague claims about "AI taking jobs."

The measurement problem is real. O'Laughlin's observation that displacement will be hard to measure — offices still full but outputting multiples more work — echoes the productivity paradox literature and identifies a genuine challenge for policymakers. Engels literally walked the streets of Manchester counting empty cottages. There is no equivalent metric for cognitive displacement.

The geopolitical dimension matters. The parallel between Britain producing half the world's iron by 1854 and the US producing the majority of the world's AI-generated code is structurally sound. National concentration of AI production capacity has redistribution consequences that cross borders — a theme explored further in The Sovereign AI Movement and AI Chip Export Controls.

Critiques and Limitations

The investor gaze naturalizes displacement. The piece's most revealing sentence is its conclusion: "Capital could have an incredible decade." O'Laughlin diagnoses a dynamic where $4.6 trillion per year shifts from workers to capital owners, describes it as a modern Engels' Pause producing a "permanent underclass," and then frames this as an investment thesis. This is not a critique of O'Laughlin specifically — it's the logic of the newsletter's audience — but it illustrates exactly the mechanism Acemoglu identifies in Paper Review: The Wrong Kind of AI?: displacement happens not because it's technologically inevitable but because it's profitable for capital allocators who treat it as a fiduciary obligation.

"Fiduciary duty" is a choice, not a law of nature. O'Laughlin presents the replacement of $120K analysts with $10K in tokens as an inevitable fiduciary imperative. But fiduciary duty is a legal and institutional construct that has been interpreted differently across eras — and can be again. The shareholder-primacy reading that O'Laughlin invokes is itself a product of the Chicago School revolution of the 1970s and 80s. Stakeholder governance, B-corp structures, and European co-determination models all offer alternative frameworks. Presenting displacement as a legal obligation obscures the policy choices that could redirect it.

The Mythos claims need scrutiny. The piece treats Mythos's capabilities as established fact and extrapolates from cybersecurity (a domain with verifiable ground truth) to the entire information economy. Finding software vulnerabilities — where correctness is binary and testable — is categorically different from producing legal strategy, financial judgment, or management decisions where context, relationships, and ambiguity dominate. The AI Benchmarks Don't Measure What You Think problem applies here: superhuman performance on one task does not imply superhuman performance on the job.

Missing: worker agency and organizing. The piece contains zero discussion of collective bargaining, labor organizing, or worker power as forces that could shape the transition. During the original Engels' Pause, the Chartist movement, the Factory Acts, and early trade unions were central to eventually ending it. O'Laughlin's framing treats workers as passive recipients of displacement — objects of economic forces rather than agents who historically fight back. The growing movement around Collective Bargaining in the Age of AI and the SAG-AFTRA/WGA strikes of 2023 suggest otherwise.

Missing: policy levers. For a piece that identifies a $4.6 trillion annual redistribution from labor to capital, the absence of policy discussion is striking. No mention of tax reform, automation taxes, portable benefits, educational investment, antitrust, or any of the mechanisms that ended the original Pause. Allen's own research emphasizes that the Pause ended through institutional and political change, not just the arrival of railroads. O'Laughlin's framing makes the outcome feel predetermined when the historical analogy itself argues otherwise.

The electrification counter-case deserves more than dismissal. David's productivity paradox research is more nuanced than O'Laughlin allows. The key variable isn't whether there's a "human in the loop" but whether the technology creates new tasks faster than it destroys old ones. Electrification created entirely new industries (consumer electronics, refrigeration, entertainment) that didn't exist before. Whether AI creates equivalent new task categories — and how quickly — is an empirical question that O'Laughlin resolves by assertion rather than evidence.

The timeline compression claim cuts both ways. O'Laughlin argues that faster technology diffusion means the Pause will unfold in years, not decades. But faster diffusion also means faster adaptation, faster policy response, and faster creation of new roles. The speed argument can't selectively apply only to destruction.

Implications for Redistribution

This piece crystallizes the redistribution question in dollar terms — and reveals how the people with capital think about it.

Value: O'Laughlin quantifies what Korinek models theoretically in Paper Review: The Distributional Effects of AI: a multi-trillion-dollar annual transfer from labor to capital. The $1.03 trillion in "at-risk compensation" across 16.7 million workers is not an abstract projection — it's a target list for capital allocation.

Work: The artisan parallel reframes the displacement debate. It's not about "low-skill vs. high-skill" — it's about which skills carry premiums that create automation incentives. This aligns with the task-based framework in Paper Review: GPTs Are GPTs and complicates the standard advice to "upskill." The handloom weavers were the skilled workers.

Power: The piece's most important contribution may be unintentional. By framing displacement as fiduciary duty and concluding that "capital could have an incredible decade," it demonstrates the mechanism by which displacement accelerates: not through technological determinism, but through the choices of capital allocators who see the Pause as an opportunity rather than a crisis. Understanding this framing is essential for anyone trying to build countervailing power.

What's missing from the redistribution picture: The piece describes the transfer but proposes nothing about redirecting it. For readers of Redistributed, the question isn't whether Engels' Pause is a useful analogy — it is — but what ended it, and what equivalent interventions look like today. Allen's own research points to institutional reform: factory legislation, union recognition, public education, and ultimately a political settlement that distributed industrial gains more broadly. The AI equivalent would include Public AI, progressive automation taxation, AI Cooperatives, and the kind of sectoral bargaining discussed in Collective Bargaining in the Age of AI.

How to Use This Piece

When explaining displacement to non-technical audiences: The Engels' Pause analogy is historically grounded and immediately intuitive. "The skilled workers with the highest wages were automated first, because the premium created the incentive" is a powerful and accurate framing.

When arguing that displacement is a choice, not destiny: Use the piece's own logic against its conclusion. If displacement is a "fiduciary duty" under current institutional arrangements, then changing those arrangements — tax incentives, governance structures, labor law — changes the calculus. The original Pause ended through political action, not technological progress alone.

When tracking where capital sees displacement opportunities: O'Laughlin's displacement incentive ranking is investor-facing analysis. It reveals where automation pressure will be applied first and hardest. Workers, unions, and policymakers can use the same framework defensively.

When countering both hype and doom: The piece avoids the extremes of "AI will solve everything" and "AI will destroy everything." The historical parallel shows that transformative technology does create enormous value — and that the distribution of that value is a political question, not a technological one.

Source

O'Laughlin, Doug. "Engels' Pause and the Permanent Underclass." Fabricated Knowledge (Substack), April 16, 2026. Fabricated Knowledge

General

Article Review: Engels' Pause and the Permanent Underclass