Bitter Losses, Sweet Gains: The Economic Prospects of AGI
Pascual Restrepo’s recent paper, “We Won’t be Missed,” offers a provocative conclusion: In a world with Artificial General Intelligence (AGI), labor’s share of income converges to zero because human labor is replaced by “compute.” This result is plausible. The modeled path to it, however, is not.
The model used in the study is a “top-down” approach. While elegant, this approach sacrifices the reality of production for the consistency of aggregation. To understand the impact of a technology as fundamental as AGI, we must switch perspectives: from macroeconomic aggregation to a “bottom-up” analysis that starts at the “atomic” processes of value creation. This is where nanoeconomics can help (Gerlach, 2025).
The Problem of Aggregation: Why Top-Down Models Fail
The model presented in the study is a perfect example of the limits of aggregation. It is questionable to analyze such a microeconomic phenomenon—the application of a new information technology at the task level—using coarse macroeconomic aggregates.
In Restrepo’s model, “Labor” and “Compute” produce an abstract, dimensionless output, which flows into an equally abstract production function. Here lies the first blind spot: In this model, no real goods—no useful products—are manufactured. The production serves an imaginary, undefined demand. How can a model evaluate the impact of a production technology if the object of production—the good created through specific processes—is not considered?
Such a model, by definition, falls short of mapping systemic relationships. It ignores the complex, networked value chains in which value is actually created. As W. Ross Ashby stated, “only variety can destroy variety.” An aggregated model that ignores the variety of real production cannot analyze a technology that specifically targets that variety.
The core error lies in the central premise: the assumption that AGI is a substitute for labor, which simply has a different cost.
What is AI’s True Contribution? (A Nanoeconomic Explanation)
To understand the causal chain of transformation, we must apply a “bottom-up” model based on the fundamental logic of economic value creation.
The Lever: AI as Delta-Info
Every single economic action follows a stable pattern that creates a causal chain: It begins with Information, which fuels a Decision, which triggers the Action that leads to a result. We call this the IDEA mechanism.
According to this chain, the only point of intervention for AI in the real economy is the very first step: Information. AI is not a direct substitute for action; it is solely an information technology. Its economic essence is that of a supplier of improved, faster, or more granular information. We call this leverage Delta-Info (∆Info).
Restrepo’s model analyzes the impact of an information technology without even having “information” or “decision” as variables. This Delta-Info is the sole lever AI applies in an economic process. All subsequent economic effects are merely the accumulated, systemic consequences triggered by this tiny, local impulse.
The Landscape: The Network of Production
These mechanisms operate within a dense, networked system. The real economy consists of millions of interconnected productive microsystems (production units). These form what we know as supply chains: The finished product of one system is the raw material for the next.
How Do Production Systems Evolve with the Advent of AGI?
If we now assume the application of AI (as a Delta-Info supplier) by economic agents within this network, something far more profound than mere substitution occurs.
1. Local Change: The AI-Fusion-Effect
Within a single production unit, the massive availability of Delta-Info through AI enables a fundamental reorganization. Any value creation process that can be safely, reliably, and efficiently automated will sooner or later be automated due to market competition.
The consequence of this striving for automated production processes is that previously separate factors of production—human control, the physical tool, and information processing—can now merge into a single, autonomous unit. We call this the AI-Fusion-Effect: the emergence of a “smart agent.” This new system is a structurally new entity that can execute the production process autonomously within defined boundaries.
Here lies the decisive difference from Restrepo’s model: The human worker does not become too expensive (their wage is not “capped”). Their function becomes structurally obsolete. The productive system no longer competes with the worker on price; it eliminates the need for their presence in the process for the products under consideration.
Since this efficiency increase will force other producers to upgrade their production systems as well, the effect will propagate—much like the butterfly’s flap that can cause a typhoon on another continent.
2. Global Transformation: Functional Scaling
A company possessing such “smart agents” is no longer constrained by old process boundaries. It must re-evaluate the classic “Make-or-Buy” decision. Why should it buy components from an external supplier when it can perform that supplier’s function itself through a simple reconfiguration of its own “smart agents”?
This is where Functional Scaling begins. Instead of negotiating prices with suppliers, the system begins to internalize their functions. This reflects the fundamental isomorphism (structural similarity) between the internal logic of a production system and the external logic of a supply chain. The visual similarity of the new, fused “smart agent” to the converging shape of a supply chain is not a coincidence. It is the internalized supply chain.
The consequence of these developments is not the optimization of the existing supply chain, but its radical shortening and even dissolution.
The Holistic View: How Do People “Buy” Goods Without a Wage?
This production-side analysis is only half the truth. Any economic analysis that stops here is incomplete. Economics ultimately serves human need satisfaction. People have needs, and the entire production sphere exists only to create goods that satisfy those needs.
This is the central blind spot that an aggregated model like Restrepo’s systematically misses: It analyzes production without ever modeling the actor who must buy the output.
Here, a central, unresolvable contradiction is revealed: To buy goods, people need a Budget. This budget historically stems from their income, which they earn through work.
What happens when—as both analyses predict—human work disappears from the production process?
- “Smart agents” eliminate the need for human work.
- Without work, income disappears, and thus the consumer’s budget.
- Without a budget, consumers can no longer buy the goods produced by the “smart agents.”
- Production collapses because its goal—whether profit or need satisfaction—is no longer met.
A system of hyper-efficient production where no one has an income is not stable. It is a system in immediate collapse. This demonstrates that only a holistic perspective—which considers production and consumption as one inseparable system—can lead to viable insights. A framework that models production but not the consumer is incomplete.
The alternative, that people could produce the goods themselves, does not initially change the dilemma: even for self-production, they need a budget to buy the material, machines, and energy.
This systemic constraint leads us to one logical conclusion: The organization of production must fundamentally change. A plausible scenario is the emergence of smart, local factories that require extremely low inputs of material and energy to produce high-quality, long-lasting goods.
These hyper-efficient production systems (the “smart agents”) can themselves become products. Thus, they can be possessed by individuals, instead of serving a global market. If we truly achieve AGI, its primary task should be to ensure that the development and replication costs for these personal smart factories converge toward zero, making the missing income irrelevant.
This techno-optimistic outlook is a modern echo of prehistory, where a family around a campfire produced everything it needed for life: food, clothing, tools, and weapons. The smart factory could enable the same self-sufficiency, but at an infinitely higher level of capability and comfort. Who knows? We could find new modern campfires.
Conclusion: Different Path, Different Policy
The final result—a drastic decline in labor’s share of income—may be the same in both models. But the path to that result is fundamentally different.
- Restrepo’s Model: A price competition (wages are “capped”).
- The Nanoeconomic Model: A structural elimination of labor from production processes (functions are ‘absorbed’), resulting in a workforce deprived of traditional wage income.
This distinction is crucial because the path determines the policy.
If you follow Restrepo’s model, the logical policy response is to make human labor competitive—through wage subsidies (to lower its price) or education (to increase its productivity).
If you follow the bottom-up analysis, such supply-side interventions are futile. You cannot compete with a system that has eliminated your function. The debate over wage caps is irrelevant when the structural necessity for the wage itself disappears. The focus must be the satisfaction of consumer needs, the “full fridge challenge.”
Could Universal Basic Income (UBI) help? Only as a temporary bridge. In the short term, UBI is essential to maintain consumer purchasing power and mitigate the structural loss of labor income. However, in the long term, UBI is mathematically unsustainable because the elimination of labor erodes the state’s tax base, while hyper-efficiency compresses the corporate margins required to fund it. True economic security will therefore not come from a permanent state allowance, but from the emancipation of production: a shift from earning wages to buy goods, to owning the “smart agents” that produce them directly.
To meet this challenge, politics has to focus on the “smart agents.” However, in an age governed by AGI, the need for human-driven politics might eventually disappear as well.
Sources:
Ashby, W. R. (1956). An introduction to cybernetics. Chapman & Hall.
Gerlach, S. (2025). The Simple Nanoeconomics of AI: An Economic World Model for Exploring AI Impacts. Available at SSRN 5394649. https://papers.ssrn.com/sol3/papers.cfm?abstract_id=5394649
Restrepo, P. (2025). We Won’t be Missed: Work and Growth in the AGI World (No. w34423). National Bureau of Economic Research. https://www.nber.org/books-and-chapters/economics-transformative-ai/we-wont-be-missed-work-and-growth-agi-world