The First Review Scored This Idea 12 Out of 100

That number is the reason you can trust what follows.

Emanuel F. Barros · July 2026 · 7 minute read

In the summer of 2026 I asked a panel of adversarial reviewers to evaluate an economic framework I had spent months building. Their instruction was simple: break it. They did. The arithmetic did not reproduce its own baseline. The incentives leaked. And the judge summarized where my design would end up in four words I have thought about every day since: "UBI plus oligopoly, eventually."

Twelve out of one hundred.

Most ideas should die at 12 out of 100. I decided this one deserved a rebuild instead, not because I was sure it was right, but because the question it tries to answer is coming for all of us whether we answer it or not.

The question nobody wants to sit with

When artificial intelligence can do essentially every job, where do paychecks come from?

The modern economy runs on a loop: wages fund consumption, consumption funds production, production funds wages. Every prior automation wave broke one link and the workforce moved to a new sector: farm to factory, factory to office. AI is different in kind, because it is a general substitute for cognition. This time the destination sectors automate too. The loop does not reroute. It severs at the wage link, while productive capacity keeps getting better.

Picture the endpoint: warehouses full, factories humming, storefronts dark, because nobody earns. Production without distribution is capacity without purpose. That is not a technology problem. It is a plumbing problem, and almost nobody is designing the plumbing.

The clutch

Here is the reframe the whole framework grew from. An engine revving with the clutch disengaged produces power, noise, and zero motion. In an automated economy, AI production is the engine, growth is the wheels, and the consumer is the clutch: the mechanism that transmits productive capacity into economic momentum.

This is more than a metaphor, because it makes a measurable claim: the distribution of purchasing power, not just its volume, determines the returns to innovation. A dollar of demand from ten million households buys more innovation than ten million dollars from one household, because breadth multiplies the viable products, niches, and experiments an innovator can sell into. Fairness, on this view, is not a tax on growth. It is fuel.

A claim like that can be written as an equation, put inside a model, and given the chance to fail. So that is what I did.

What I built after the 12

I rebuilt the framework inside a simulated economy: 250 million households, running from 2025 to 2075, each with realistic starting wealth and widely varying ability. The critical design choice is that innovation is endogenous. Nobody hands the economy a growth rate. Simulated people launch ventures only when the expected prize justifies the attempt, which means any policy that shrinks prizes or blocks capital access shows up as ventures that never happen. Every anti-innovation failure mode critics worry about is free to appear in the results.

Then I did the part that I now consider non-negotiable for anyone proposing an economic design: I hired the meanest reviewers I could construct, three separate times, and every exploit they found was programmed into the simulation as an attack before any rule was patched. A patch had to beat the executed attack, not the described one.

Three times, the model overruled me. I wanted short innovation rents; the model showed long ones grow the economy twelvefold with barely any inequality cost, so creators keep their winnings for decades. I designed a prize mechanism I was proud of; the model showed it bought zero additional innovation and only concentration, so it died. And the model caught my own favorite real-income number resting on a frozen consumption basket, so the framework now reports the honest pair instead. An idea that can lose arguments with its own evidence is an idea whose wins mean something.

What the model says

The design that emerged rests on ownership, not welfare. Every citizen holds one non-transferable license for a personal AI, a Digital Twin, that earns market income on their behalf, on hardware they own. A citizen-owned fund, executed by open-source code that no politician or fund manager can steer, pays a universal dividend. Innovators keep outsized returns for decades, and only passive returns on large accumulated fortunes are progressively socialized. A guaranteed floor sits underneath.

The central result, and the reason this framework exists:

The fair architecture out-grows laissez-faire at every demand-feedback strength tested. Not "grows almost as fast." Out-grows, from the weakest feedback setting to the strongest.

At the empirically anchored calibration, the simulated 2075 economy reaches $875 trillion in GDP with a median household income of $1.69 million, 23 times today's. Income concentration falls below today's level, no household sits below the floor, and new ventures launch at rates above today's baseline. The wealthy do not lose: the top 1% hold a somewhat smaller share of a pie so much larger that they end up roughly 25 times richer in absolute terms. Nobody is leveled down. The zero-sum instinct that says the rich must lose for everyone else to gain is precisely the scarcity-era reflex this result retires.

One more result, my favorite, from the final red-team round. When the strongest combined attack its adversaries could compose was run against the finished design, GDP and the universal dividend went up, because the design's exposure tax converts attack capital into everyone's dividend. The attackers end up funding the public. The reviewers' score went from 12 to 22 to 65 to 67 across the rounds, and every report is published.

What I am not claiming

These are scenario analyses, not forecasts. The key empirical parameter is anchored in cross-country data with the weak spot disclosed: within-country evidence cannot yet reject zero. The model is a closed U.S. economy. Three rounds of adversaries plus an optimizer bound the attack space; a genuinely novel exploit sits outside it. The code is open because I want it broken by someone smarter than my red teams. Every exploit found gets encoded, tested, and published with credit.

Neither tribe's project

Readers reach for familiar boxes, so let me pre-empt the two most common. This is not socialism. Private property runs all the way down: your twin, your hardware, your businesses, and everything the twin earns is fully yours to sell, invest, or leave to your children. Nothing is confiscated, existing wealth is untouched, and allocation happens through markets, not plans. What gets universalized is ownership and access, not outcomes. Think of the fund as shareholder capitalism completed: every citizen holds shares in the automated base, under the same public rules, with advantages for none. If universal share ownership is socialism, then so are index funds.

And it is not a defense of the status quo, because the status quo has no answer to the severed loop. Concentrated ownership of automated production is not capitalism's triumph; it is capitalism running out of customers. The design's bet is that both instincts, the one that prizes growth and merit and the one that prizes security and inclusion, are satisfiable by the same architecture, because in a post-job economy fairness and growth stop being rivals. The framework belongs to neither party, and it will stay that way.

You can start before the law does

Here is what makes this framework different from every other future-of-work proposal I know: Phase 0 requires no legislation, no political victory, and no one's permission.

An income-earning AI agent operated under ordinary business law, what the framework calls a proto-twin, is legal today. Agentic freelance services, automated e-commerce, content operations, AI consulting: individuals are already running them. Every honest proto-twin generates two things the transition needs, income data and a constituency, long before any licensing debate begins. Adoption leads. Policy follows. The people deploying agents this decade are writing the norms the statutes will later codify, which is exactly why the protections need to arrive with the first licenses rather than after the aggregators entrench.

The machines should buy us mornings. Whether they buy them for everyone, or for whoever owns the datacenters, is not a technology question. It is a design question, and the design window is open right now.

I am not asking you to believe me. I am asking you to check.