Suppose robots reduce a factory's full cost per accepted product by US$100.
Customers might pay less, shareholders earn more, and remaining employees receive bonuses—or lose income as overtime falls. Displaced workers face the time and risk of finding another job.
The same technological advance can produce all these outcomes.
We have spent plenty of time discussing how much more machines can do. We also need to ask:Through what channels does additional productivity improve ordinary people's income and lives?
Saving US$100 does not tell us who gets it
Full cost here includes depreciation, financing, maintenance, energy, human support, and retries. Comparing robot electricity bills with wages would not even establish the size of the gain.
Even with a confirmed US$100 saving per item, distribution is not automatic.
In a simplified case with sales and other conditions unchanged, a US$60 price cut leaves US$40 for compensation, retained profits, or other spending. If competition passes the whole saving to customers, buyers benefit directly. If prices stay unchanged, more remains in the business.
Taxes then affect what firms and individuals retain, while spending and transfers affect others. Price cuts, wages, pretax profits, and taxes cannot be drawn as four independent slices of one pie: that would double-count.
Nor should we assume all gains become dividends. A business may need to repair its balance sheet, expand capacity, or fund risky research. The questions are how those decisions are made, who shares their long-term benefits, and who bears their costs.
Technology makes it possible; markets and bargaining divide the gains
Identical robots can produce very different distributional results in different markets.
When customers can switch suppliers easily, savings are more likely to lower prices. When a few firms control interfaces, data, or distribution, savings may raise profits. Scarce deployment, maintenance, and exception-handling skills may command higher pay; workers with few alternative employers may see much smaller raises than productivity gains.
Which of these mechanisms matters most will vary from market to market.
The ILO's generative AI research examines tasks and distinguishes occupational transformation from wholesale replacement. Jobs contain multiple tasks; when AI takes some over, the value and organization of the remaining work still change. Research on generative AI is not direct evidence that humanoid robots already have equivalent capabilities.ILO research overview
When a business says AI improved employee efficiency, it should also explain what followed: higher income, shorter hours, more orders, or tougher performance targets?
Those four outcomes feel very different to employees.
Narrower wage gaps can coexist with wider wealth gaps
Suppose AI helps ordinary workers perform tasks previously reserved for specialists. Wage gaps could narrow—a welcome possibility.
But household resources include more than wages. Equity, business ownership, and other investment income also shape long-term prospects.
A 2025 IMF working paper uses household microdata to calibrate a task model examining this divergence. Under some conditions, AI substitution in high-income jobs can narrow wage inequality while capital returns and adoption decisions widen wealth inequality. This is model-based analysis, not an observed survey of a robotic society.IMF：AI Adoption and Inequality
The implication is specific:To judge whether AI's benefits are broadly shared, examine wages, capital ownership, and who receives asset returns. Those returns may come through dividends, interest, or changes in asset prices; price appreciation is not a prerequisite for sharing the gains.
Within one robot service network, one person buys services, another earns operating wages, and another owns platform equity. All participate, with different income sources and risks.
Proken, introduced earlier, is a proposal for measuring and trading work. Easier service access would not automatically change asset ownership. Access and ownership require separate arrangements.
The strongest counterargument: lower prices are a form of distribution
A strong counterargument is that even concentrated profits can accompany widespread benefits if services become much cheaper. Cash receipts are not the only benefit; purchasing power matters too.
That argument is valid. Cheaper cleaning, repairs, and delivery could reach households that previously could not afford them. Cheaper productive inputs could also let small businesses enter markets, expanding employment and competition.
But the net benefit must be assessed household by household.
Someone who saves US$300 a month on services but loses US$3,000 in wages may still be worse off. Someone with secure work and assets may enjoy both lower prices and investment returns. These numbers illustrate a mechanism, not a forecast.
Whether housing, care, and energy also get cheaper depends on their supply constraints. Near-free digital services do not imply that every living cost will disappear.
Price benefits, income changes, and transition costs must be counted together to identify who actually becomes better off.
Companies could start by reporting how gains are shared
I would like to see more complete AI deployment reports.
They would disclose not only costs per unit of accepted work, but changes in prices, actual employee pay and hours, internal transfers, hiring, investment recovery, and profits.
This does not prescribe one sharing ratio. It makes claims of large efficiency gains verifiable. Appropriate choices vary by industry, place, and deployment stage, but reports should show both benefits and costs.
Policy needs similar care. Competition, transition support, taxes, public services, and wider access to asset ownership address different problems. Telling everyone to learn AI cannot resolve every ownership and bargaining issue.
Physical AI could reduce the resources needed for many tasks. Whether that improves lives broadly depends on how wages, prices, capital, and public institutions are organized.
Productivity growth creates room to share. People must still decide how.
The next essay asks whether, beyond distributing gains after they arise, more people could own the assets that produce them.
When your organization shares efficiency gains, what changes first: prices, pay, hours, or profits?