The Economy May Not Need Our Labour. Who Gets to Own It?

Why this question feels urgent

On 26 October 2026, How To Prompt posted on X about Pascual Restrepo's AGI paper. The post opened with a claim that turns an abstract economic model into a direct threat to people's livelihoods:

“Researchers mathematically proved that AI will destroy the middle class.”

That framing makes the question immediately personal. A job can provide income, routine, recognition and a place in society, so the prospect of machines performing our work raises questions about how we will live and where we will belong.

The paper makes a more specific argument: under its long-run assumptions, labour's share of GDP can approach zero while people still earn positive wages. A mathematical result within those assumptions is not proof that the middle class must disappear, or that half of today's workforce could stop tomorrow without reducing output.

I think the ownership question deserves attention even without that certainty. If an economy can become increasingly productive without depending on more human labour, how do people gain a share of its prosperity and a voice in its direction?

To connect that question to ordinary life, imagine a day spent doing things that matter. You help someone understand an idea, build something useful, prepare food, exercise, resolve a disagreement and recover from exhaustion. Some of these activities generate income, while others support lives and relationships without producing a transaction.

Now imagine machines performing an increasing share of the activities for which people currently receive wages. We would still need resources, relationships, recognition and opportunities to shape our surroundings. The question becomes how institutions provide these things when employment changes.

Quick Answer

AI-driven growth could increase output without automatically giving people ownership, income or influence. This essay uses one founder experience to examine wellbeing and contribution, compares two income designs that remain proposals, and asks what evidence would be needed before a score could affect public payments.

What changes when machines can do the work?

Pascual Restrepo's We Won't Be Missed examines a theoretical economy where all economically useful work can be performed using compute and computational resources keep expanding. Under its assumptions, growth-critical work is eventually automated; people can retain jobs and positive wages while labour's share of income approaches zero. This is a conditional result, rather than an arrival date or a prediction that everyone becomes poor.

The author-hosted revision dated 22 October 2025 is later than the July conference draft that prompted this essay. It uses “supplementary work” in place of “accessory work.” The distinction concerns an activity's relationship to sustained growth; it does not settle its importance to a person's life. Read the revised chapter.

For illustration, an economy could grow from 100 to 1,000 while its wage bill rises from 60 to 80. Workers receive more in absolute terms, but their share falls from 60 percent to 8 percent. These invented numbers illustrate the difference between an income level and an income share; they are not a forecast.

The institutional question I take from this is how people gain meaningful access to an increasingly productive economy. A society might increase output without giving everyone a corresponding share of its returns or authority over its direction. This connects to my work on The Internet of Value, while challenging us to explain precisely which parts of the problem our work addresses.

A real day from my Value Log

Part of this work begins with an ordinary day. On 25 October 2026, my Value Log listed 16 activities covering 22 hours and 45 minutes. They included exercise, meals, conversations, product development and an evening awareness exercise with other participants.

During that evening exercise, I invited people to mentally replay their day and consider their experiences through Physiology, Emotion, Feeling, Thought, Habit and Performance. The exercise itself did not require phones or logging. The captured view shows how I subsequently represented that activity within my own record.

This image demonstrates a working representation of activity. It lets a reader inspect how time, description, interpretation and review status appear together. It does not establish that every description is accurate, that the day was completely captured or that the activity caused an improvement in wellbeing.

Those limits are part of the experiment. Before a record supports a consequential decision, the person represented needs a practical way to inspect it, correct it and challenge its interpretation. Missing information must remain distinguishable from an absence of contribution.

The experience is also a founder example. It cannot establish how reliably the product works for other people, devices or circumstances; that requires broader testing. Our recent real-person logging difficulties make this an immediate responsibility rather than an abstract qualification.

What my identity and score can show

A day record captures selected activity, but a person also carries a history, values, capabilities, relationships and an imagined future. Employment describes only part of that experience. This is where Wellbeing Identity enters the work.

The protocol describes ~WellbeingIdentity as a living state container, updated through the wider protocol system and individual authorship. It distinguishes aspects such as Story, Skills, Moral Compass, Identity State, and Consent and Disclosure. Unknown and withheld information are valid states, and the representation does not claim to exhaust the person. Explore the protocol registry.

The practical ambition is continuity with agency. Someone should be able to describe developing skills, changing circumstances or a new direction without being confined to their most recent job title. They should also control which parts of that account become available in a particular relationship.

My WellbeingScore introduces a further layer of interpretation. The published view shows 771.63 out of 1,000, calculated under HouseOfBros Founder Policy v0.1, with a displayed calculation time of 26 October 2026 at 00:15 IST. This is a dated projection, rather than a continuously current reading.

Showing this number makes the implementation available for scrutiny. Readers can ask what entered the calculation, how the rules work and which interpretations are justified. The displayed decimal places describe computational output; they do not independently establish measurement accuracy.

That distinction becomes especially important when considering government use. A personal founder experiment cannot, by itself, establish that a score is suitable for determining another person's income. The six private components are not disclosed by this public projection and must not be inferred from its composite value.

What do the three images actually establish?

The day image demonstrates a particular recording interface and retains its review status. The Identity image demonstrates a public representation and a disclosure boundary. The score image demonstrates a dated output under an experimental issuer policy. Together they make one founder experience inspectable; they do not establish scientific validity, wider product reliability, economic rights or policy effectiveness.

What newer research adds

The research conversation has moved beyond a single AGI thought experiment. It now includes evidence about current usage, studies of labour-market outcomes, models of possible transitions and proposals for distributing economic returns. These sources address different questions, so their evidence types need to remain visible.

These sources strengthen the case for investigating institutions while preserving uncertainty about the scale and timing of change. They do not provide evidence validating our wellbeing score or demonstrating an IoV income mechanism. Provider usage data also describes selected products and users, rather than the entire economy.

Someone might use a powerful AI tool every day while holding no ownership claim over the productive assets behind it. Access to tools and participation in their economic returns therefore deserve separate examination. The next interactive diagram follows the institutional connections that would have to be established.

From universal basic income to universal high income

Universal basic income provides one possible foundation for economic security. The World Bank describes UBI as a cash transfer provided universally and unconditionally; making payment depend on a wellbeing target would change that defining feature. Its policy guide also emphasises financing and system-wide questions that small pilots cannot fully resolve. Read the World Bank guide.

For this discussion, I use universal high income to describe an ambition for broadly shared purchasing power substantially beyond essential needs. That ambition requires a clear definition of “high,” a sustainable source of resources and attention to what people can actually obtain with the money. The phrase is an objective to specify, rather than a funding mechanism.

A larger nominal payment would provide limited progress if housing, care, energy or other necessities became correspondingly harder to access. Income design therefore needs to be considered alongside productive capacity, public services and the institutions governing scarce resources. It also needs to account for the effects of changing existing disability support and other targeted provision.

Ownership could contribute to the funding side. OpenAI's public wealth fund proposal envisages citizens receiving returns from a diversified pool of productive assets. It is an exploratory proposal, rather than evidence that a particular dividend level is already affordable.

A wellbeing score does not create the resources required for such a system. Its possible contribution lies in helping people and institutions understand outcomes, identify problems or organise a carefully bounded incentive programme. Its usefulness needs to be demonstrated for each of those purposes.

Two ways wellbeing could enter income policy

We are considering two designs, and they make different promises. The first protects an unconditional income floor while allowing separate experiments; the second uses wellbeing to determine how much someone receives. Neither has been adopted as an IoV public-payment policy.

Design A: protect the base and examine additional incentives

Under this design, a person's basic payment would not fall because their score declined or they stopped logging. A separate programme might offer voluntary goals, feedback or capped additional payments under explicit rules. The purpose and funding of that additional programme would need to be explained independently.

That separation does not settle every concern. If the floor is inadequate or the additional payment is large enough to become essential, participation may remain difficult to refuse. Accessibility, administrative burden and alternative routes need to be evaluated alongside nominal consent.

Design B: let wellbeing affect the payment itself

A score-linked design might appear to make public spending more responsive to outcomes. It could offer participants a visible payment rule and give administrators an explicit mechanism to examine. Those features would be useful only if the underlying measure and allocation principle were defensible.

The difficult question is what a low score should mean. It might indicate circumstances requiring more support, while a reward formula could give the same person less money. Measurement cannot resolve that normative choice.

We would also need to distinguish among several possible rules. Their consequences cannot be settled by calling them all wellbeing incentives.

• Rewarding a high state could advantage people whose circumstances already make wellbeing easier to sustain. A high reading would need an appropriate interpretation for the decision.
• Rewarding improvement could disadvantage people maintaining a stable state and create incentives around establishing a low baseline. The comparison window and baseline would become consequential choices.
• Rewarding participation would measure engagement with a programme rather than improvement in wellbeing. The distinction should be explicit in both the payment rule and its evaluation.
• Responding to need would pursue a different objective from rewarding performance. A need-responsive allocation should be examined on its own terms.

Some of our own SAO design materials contemplate wellbeing as a funding signal. Those provisions deserve the same examination before being extended into government-income proposals. Existing design language should remain open to challenge rather than being treated as evidence of policy effectiveness.

There is also a boundary around universality. A high income achieved only by people who meet a score condition would not fulfil the universal ambition for those who cannot meet it. We would need to state which part of the income is guaranteed and which part remains conditional.

What gamification evidence actually supports

There is empirical support for investigating particular combinations of goals, feedback, gamification and financial incentives. The BE ACTIVE randomised trial, published in 2024, studied 1,062 adults at elevated cardiovascular risk and found increased daily steps in its intervention groups compared with attention control. The combined gamification-and-incentive group showed the largest increase during the intervention. Read the trial.

That result supports a bounded claim about a defined intervention, population and outcome. It does not establish that a composite wellbeing score should determine income or that the same approach would improve every aspect of life. For our work, the question is whether a specific programme helps particular people under stated conditions.

Cash-transfer evidence also cautions against assuming a simple relationship between money and wellbeing. OpenResearch's targeted US experiment provided 1,000 participants with $1,000 monthly for three years, while 2,000 comparison participants received $50 monthly. It was an unconditional cash experiment, rather than a nationwide UBI programme, and OpenResearch is a separate organisation from OpenAI. Study design.

Its health findings reported first-year improvements in stress, mental distress and food security that subsequently faded, alongside increased use of some healthcare services and no significant measured physical-health improvement. These findings concern that programme and context; they demonstrate why we should examine multiple outcomes over time. Health findings.

The experiment worth conducting

A useful research programme would separate usability, measurement, incentives and institutional effects. My personal example can help examine the first of these, while much stronger evidence would be required for public allocation decisions. The sequence matters because an interpretable interface and an appropriate income rule are different achievements.

First, establish a trustworthy personal record

People should be able to recognise their experience in the record, correct errors and retain uncertainty. Self-report, personal confirmation, community review and scientific validation should remain distinguishable. A small pilot could examine these practical capabilities, but could not establish national fiscal feasibility or a general improvement in wellbeing.

Second, test the measure for its intended purpose

A score intended for private reflection faces different requirements from one intended to influence payments. We would need to define what it represents, examine reliability and interpretation, and understand how results vary across circumstances. The current founder projection does not supply this evidence for government use.

Third, distinguish additional money from incentives

If a programme provides both additional money and gamification, an improvement alone would not tell us which part helped. A credible comparison could include the following conditions, subject to appropriate research design and participant protections.

1. An unconditional base with ordinary support, providing a reference condition.
2. The same base with voluntary goals and feedback, examining the additional contribution of that experience.
3. The base, feedback and a capped incentive, examining the combined intervention.
4. A comparable unconditional additional payment, helping distinguish incentive design from simply receiving more money.

The exact study would require appropriate expertise and sufficient scale. Outcomes should include independently assessed hardship, autonomy, burden and relevant wellbeing domains. Evaluating success only through the score being rewarded would leave the central question unanswered.

Fourth, establish the institution behind the promise

Any income mechanism needs an identifiable source of funding, understandable rights and a workable process for correction and appeal. It also needs a plan for service failures, interrupted participation, changing circumstances and disagreement. Participants should be able to explain the commitment in ordinary language.

The current protocol boundary distinguishes final valuation from payment: its status is non_payable, rather than settlement or a transferable balance. A government-income mechanism would therefore be additional institutional work, not a capability demonstrated by the existing founder score. Read the public Value Log method.

What shared prosperity should make possible

The possibility that interests me is an economy that gives people more room to author their lives. Someone might use greater security to learn, care for another person, recover from exhaustion, build a community or leave work that is harming them. These possibilities deserve attention alongside changes in output.

Understanding those outcomes could make a wellbeing framework useful. Its usefulness would come from supporting better judgment and decisions, with evidence appropriate to the purpose. Attaching money to a number makes it consequential; it does not automatically make it a better measure.

This leaves us with two responsibilities. We need institutions through which people can receive and influence the returns of productive systems, and we need ways to examine what those arrangements do to their lives. Neither responsibility can be completed by a screenshot, a contribution ledger or a single founder's experience.

The images in this essay make a small part of our work inspectable. The unresolved questions remain substantial, including whether income-linked incentives help, whom they disadvantage and whether the administrative system respects the agency it is intended to expand. Those questions should be able to change the design.

I want us to build toward a society where economic security gives people more freedom to participate, rather than making every experience another test of eligibility. If machines can do more of the work, people should have a greater opportunity to live well and a meaningful voice in deciding what that means.

Frequently Asked Questions

Does AI automatically make universal high income affordable?

No funding level is established by the argument or the interactive illustration. A universal income commitment needs actual resources, a definition of adequacy, and attention to prices, public services and existing support.

Does the founder score measure everyone's wellbeing?

The displayed number is a dated, experimental projection under HouseOfBros Founder Policy v0.1. It does not establish a universal human ranking, clinical assessment or government eligibility measure.

Does a contribution record create a payment entitlement?

A record can inform review and explicit institutional agreements. The current final-valuation boundary is non-payable; money and economic rights require further funded arrangements.

Has either income design been adopted?

Both remain proposals under examination. The comparison is a local mathematical illustration and does not change a payment, personal record or eligibility decision.

References

The research linked throughout this essay was checked on 27 October 2026. The personal images are implementation evidence, while the income designs and study comparisons are proposals. The distinction between these evidence types is part of the argument.

The opening quote comes from How To Prompt's 26 October 2026 post. It is cited as context for the public discussion; the research claims are checked against the papers below.

Economic models, observed use and policy proposals

• Restrepo's revised AGI chapter, 22 October 2025: a conditional long-run model.
• Hassan, Kalyani and Restrepo, March 2026: technology creation, standardisation and the skill premium.
• Lehr and Restrepo's public IMF working paper, November 2025: socially minded AI pricing and deployment.
• Anthropic's March 2026 labour-market analysis: observed exposure and early employment evidence.
• Anthropic Institute's September 2026 scenarios: conditional paths, with no assigned probabilities.
• OpenAI's 26 October 2026 data release: consumer work-related usage, excluding enterprise and Codex.
• OpenAI's April 2026 policy paper: exploratory proposals, including a public wealth fund.

Income, incentives and the current implementation

• World Bank guide to UBI: definitions and system-wide design questions.
• OpenResearch study design and health findings: a targeted US cash-transfer experiment.
• BE ACTIVE trial: gamification and financial incentives for a specific physical-activity outcome.
• My owner-approved public Identity and Value Log method: the current example and calculation context.
• IoV protocol registry: definitions and the four-protocol system. The event-level ProtocolLinkId carries lineage; it does not establish a financial entitlement.