When AI takes the last job, who owns what comes next?

Clutch Economics is a complete, quantitatively tested framework for the automated era: every citizen owns one licensed AI that earns on their behalf, a citizen-owned fund pays a universal dividend, and innovators keep outsized rewards for decades. The result, in a quarter-billion-household simulation: the fair design out-grows laissez-faire at every feedback strength tested.

New here? Read the 18-page primer Economist? Read the research paper
$875TProjected U.S. GDP by 2075 (agent-based median, anchored calibration)
23xMedian household income growth: $73K to $1.69M
0.437Income Gini in 2075: less concentrated than today
0Households below the guaranteed floor, in every tested configuration

Epistemic status, stated plainly: these are scenario analyses, not forecasts. Every number is computationally verified, the model code is open, and the framework's weaknesses are published alongside its results.

The first review scored this idea 12 out of 100.
We published that too. Three adversarial red teams later, the design out-scores its own hardened core.
12 → 22 → 65 → 38 → 55 → 67
How it survived ↓

Follow the work as it happens

New papers, model updates, and red-team results, delivered as they publish. Every claim checkable, every weakness disclosed, and the next attack round announced to subscribers first. No spam, unsubscribe anytime.

The Idea

The consumer is the clutch

An engine revving with the clutch disengaged produces power, noise, and zero motion. In an automated economy, AI is the engine and consumers are the clutch: the mechanism that transmits productive capacity into economic momentum. No broad purchasing power, no growth, no matter how good the machines get.

The problem

The wage loop breaks

Wages fund consumption, consumption funds production, production funds wages. AI severs the loop at the wage link while leaving productive capacity intact. Past automation moved workers between sectors; a general substitute for cognition automates the destination sectors too.

The insight

Breadth is a growth engine

A dollar of demand from ten million households buys more innovation than ten million dollars from one household. Broad purchasing power multiplies viable products and niches, so bigger prizes pull more ventures into existence. Distribution is not charity; it is fuel.

The test

A metaphor you can regress

The clutch is formalized as demand breadth in a growth equation, anchored in cross-country data (a 146-country regression), and tested in an agent-based economy where innovation is endogenous. If it were wrong, the model could have said so. It said the opposite.

The Design

One license, three income streams, one floor

Not UBI. An ownership architecture: the earning instrument can never concentrate, while everything it earns is fully private property.

Channel 1: primary

Digital Twin income

Every citizen holds exactly one non-transferable license for a personal AI agent that earns market income on their behalf, on hardware they own. Half the licensed pool is split equally per license; half is earned by how well you direct your twin. Ambition is never capped: twins scale into sub-agent enterprises of any size, taxed progressively like any business.

Channel 2: secondary

The universal dividend

A citizen-owned fund, executed by AI under a fixed, transparent, open-source mandate no politician or manager can steer, pays every citizen an equal dividend from commoditized profits and socialized passive returns. Shareholder capitalism universalized: capitalism completed, not socialism introduced.

Channel 3: merit upside

Innovation rents

Creators keep outsized returns for decades, with protection that fades only as products commoditize (25–40 year half-lives). Innovators are never clawed back; only passive returns on large accumulated wealth are progressively socialized. Wealth already compounds on its own. Creation is what needs the help.

Beneath everything

A guaranteed floor

About $40K (2025 dollars), honestly indexed to what people actually buy, topped up by the fund whenever the three channels fall short. In the simulations, top-ups are needed only during the transition; by mid-century the floor binds for no one, and no household falls below it in any tested configuration.

The moral core

Creators over holders

Inequality persists above the floor, by design: a super-innovator should out-earn someone who prefers to consume. What the design forbids is aristocracy by accumulation. Every anti-gaming rule, from anti-peonage caps to the exposure tax, exists to keep the earning instrument in citizens' hands.

The Evidence

Fairness out-grows laissez-faire

The central result. In an agent-based economy of 250 million households where agents launch ventures only when the prize justifies it, the fair architecture beats concentration at every demand-feedback strength tested. Two mechanisms laissez-faire structurally lacks: bigger prizes (broad demand) and more innovators (universal capital access).

Feedback strength κGDP 2075: this designGDP 2075: laissez-faireGini: this designGini: laissez-faire
0.10 (weak)$144T$117T0.420.70
0.20$306T$193T0.390.73
0.35$786T$351T0.360.77
0.50 (strong)$1,881T$527T0.340.79

Monte Carlo medians. The feedback strength κ is empirically anchored at 0.37 by literature triangulation and a fresh 146-country regression (p = 0.004), with the disclosed caveat that within-country estimation cannot reject zero. The design wins even at the weakest setting.

Nobody is leveled down

The top 1% hold a smaller share (26% to 19%) of a pie roughly 35 times larger, ending about 25 times richer in absolute terms while the median grows 23-fold. The zero-sum instinct that the rich must lose for others to gain is precisely the scarcity-era thinking this result retires.

Stronger under the critics' assumptions

Run the model with the mainstream view that capital's share swells as automation completes, and the design performs better, not worse: the swollen profit pool feeds the fund, and the dividend automatically replaces thinning twin income. The architecture contains a built-in stabilizer against exactly the shift the literature predicts.

Innovation guard: always on

Every experiment reports venture launch rates. Any configuration below 90% of baseline fails, whatever its other virtues. The headline design runs at roughly 111% of the guard threshold: this economy launches more ventures than today's, not fewer.

The model overruled its designer. Three times.

Long innovation rents beat the author's preferred short ones. A prize mechanism the author liked bought zero innovation and was deleted. The paper's most flattering real-income number depended on a frozen consumption basket, so both honest numbers are now always reported (3.3x against frozen weights, 1.54x against evolving weights). A framework that can lose arguments with its own evidence is a framework whose wins mean something.

Adversarial Testing

Attacked three times before publication

Five adversarial personas (a regulatory-arbitrage attorney, a platform monopolist, a financial engineer, a political-capture lobbyist, a shadow economist) plus an independent judge, three separate rounds. Every exploit they found was encoded in the model as an adaptive attack before any rule was patched.

Round one: 22 → 65

Buying the future

Four of five attackers converged on the same exploit: the twin is non-transferable, but buying its output stream forever buys the twin. Unpatched, the economy decays into oligopoly (Gini 0.73, top 1% holding 45%). Patched with anti-peonage caps and advisory triggers, capture falls from 36% of twin income to 5%.

Round two: 38 → 55

The fleet rebuilt

New mechanisms attracted new attacks: synthetic ownership through contingent financing, and a "putting-out system" that bought twin output at fixed prices and broke the innovation guard outright. The lesson: stop patching instruments and tax economic substance itself, whatever the paperwork says.

Round three: 67

The laundry and the pipeline

The final round attacked the fix: offshore laundering (the worst single channel found, draining 35% of the twin pool), matched-book derivatives, manufactured monopsony, and all of it composed into one pipeline. The final amendments contain everything: independence holds at 83%, leakage falls to 6%.

2075 outcomeNo attackStrongest composed attack
Real GDP$875T$954T
Median income$1.69M$1.76M
Universal dividend$159K$229K
Innovation guardPASSPASS

The design is antifragile to its strongest known attack: under the full composed assault, GDP and the dividend rise, because the exposure tax converts attack capital into fund income and venture financing. The attackers end up funding everyone's dividend.

What we do not claim

The key parameter's anchoring is cross-country; within-country estimation cannot reject zero. The model is a closed U.S. economy. The attack space is bounded by fifteen persona-passes plus an optimizer; a genuinely novel exploit class would sit outside it. Attack magnitudes are structured estimates, so results verify relative mechanism performance, not absolute damages. The open code below is a standing invitation to find what three rounds missed.

The Roadmap

Phase 0 needs no one's permission

The transition's riskiest stretch is its beginning, so the design leans on the Digital Twin first: individuals can deploy income-earning AI agents today, under existing law, building capability and evidence long before policy catches up. Everything Phase 0 needs already exists.

Phase 0 · 2026–2030

Bootstrap

Proto-twins under ordinary business law: agentic freelance services, e-commerce, automated consulting. Open-source frameworks, voluntary attestation, early twin cooperatives. Adoption leads; policy follows.

Phase 1 · 2026–2040

Foundation

Twin licensing enacted with its protections built in from day one. Displacement-triggered income support at $30K scales with actual automation. Stress-tested: even with automation seven years early, the system bends without breaking.

Phase 2 · 2040–2060

Scaling

Full conventional automation. The floor reaches $150K equivalent. The citizen fund matures as first-generation products commoditize. Consumer spending begins directly channeling startup formation.

Phase 3 · 2060–2075

Abundance

Advanced manufacturing collapses the cost of most goods. Income levels become increasingly symbolic as prices fall. Consumers direct innovation through spending and priority prizes. The engine of the economy is, explicitly, everyone.

From the Blog

The First Review Scored This Idea 12 Out of 100

That number is the reason you can trust what follows: the origin of Clutch Economics, the model that overruled its designer, and why you can start today. 7 minute read.

Read the post All posts
The Library

Free to read. Built to share.

No gates, no signups. Pick the document that matches you, and if it convinces you, send it to someone.

Start here · for everyone

Clutch Economics, Explained for Everyone

The full framework in plain language: the clutch, the license, the fund, every equation translated, every red-team round told honestly. No economics background needed.

Download PDF18 pages
For the curious skeptic

The Peer-Review Paper, Explained for Everyone

What peer review is, what the research paper claims, how the evidence works, and where it is weakest, including the parts we flag ourselves. The honesty tour.

Download PDF16 pages
For the numerate

The Math of Clutch Economics

Every equation and variable of both papers, taught with worked examples you can verify on one sheet of paper: compounding, growth accounting, Gini, the growth equation, the regression, and the design's own rule arithmetic.

Download PDF18 pages
For economists

Demand Breadth and Innovation-Driven Growth in a Highly Automated Economy

The peer-review companion paper: the mechanism formalized, the agent-based analysis, the empirical anchoring with its limitations stated prominently, and the strategic-manipulation robustness results. Written to journal conventions.

Download PDF33 pages
The source of record

Clutch Economics (framework paper, v6.1)

The complete architecture with every number: the three-channel design, the macroeconomic scenarios, the agent-based results, all three red-team rounds, the transition roadmap, and the limitations. The document everything else on this page summarizes. Companion materials include the full model code (v2 through v7), council reports, and the regression data and code, available with the paper.

Download PDF28 pages
Get Involved

Check the work. Then spread it.

Three ways to help right now

  • Share a primer. Most people first meet this idea through fear of AI and jobs. Hand them the 18-page primer instead: a future where job loss means ownership, not poverty. Copy this site's link
  • Break the model. Researchers: the code, the councils' reports, and the regression package are open. Exploits found will be encoded, tested, and published with credit. That is how every improvement so far has happened.
  • Start a proto-twin. Phase 0 is legal today: an AI agent doing real freelance work under ordinary business law. Every honest proto-twin is evidence, and evidence is what wins the policy debate later.

Questions, critiques, collaborations: eman@clutcheconomics.com

Follow the work

New papers, model updates, and red-team results as they happen. No spam, unsubscribe anytime.