Post-AGI Economies: Autonomy and the First Fundamental Theorem of Welfare Economics

Abstract: The First Fundamental Theorem of Welfare Economics assumes that welfare-bearing agents are autonomous and implicitly relies on a binary distinction between autonomy and instrumentality. Welfare subjects are those who have autonomy and therefore the capacity to choose and enter into utility comparisons, while everything else does not. In post-AGI economies this presupposition becomes nontrivial because artificial systems may exhibit varying degrees of autonomy, functioning as tools, delegates, strategic market actors, manipulators of choice environments, or possible welfare subjects. We argue that the theorem ought to be subject to an autonomy qualification where the impact of these changes in autonomy assumptions is incorporated. Using a minimal general-equilibrium model with autonomy-conditioned welfare, welfare-status assignment, delegation accounting, and verification institutions, we set out conditions for which autonomy-complete competitive equilibrium is autonomy-Pareto efficient. The classical theorem is recovered as the low-autonomy limit.

• The paper presents a novel autonomy-qualified restatement of the First Welfare Theorem by integrating human and AGI autonomy into competitive market frameworks.
• The methodology extends the Arrow–Debreu model to include detailed constructs for autonomy, delegation misalignment, and institutional governance via pricing and rights assignment.
• The findings identify critical failure modes—delegation, externality, and verification failures—that can compromise market efficiency when autonomy margins are ungoverned.

Autonomy-Qualified Restatement of the First Welfare Theorem in Post-AGI Economies

Introduction

The paper "Post-AGI Economies: Autonomy and the First Fundamental Theorem of Welfare Economics" (2604.21216) presents a rigorous restatement of the First Fundamental Theorem of Welfare Economics (FWT), incorporating the distinctive forms of autonomy introduced by Autonomously Acting General Intelligence (AGI) systems. The argument advances autonomy from an ambient background feature into a central organizing primitive in welfare economics. The result is a set of sufficient institutional and allocative conditions under which decentralized competitive equilibrium retains welfare-theoretic significance in economies where both human and artificial actors participate with heterogeneous, explicitly-modeled autonomy rights and delegation structures.

Model Structure and Key Definitions

The model builds on the Arrow–Debreu general equilibrium framework, extending it via an explicit partition of entities into welfare-relevant types—humans, tools, delegates, agentic AIs, and artificial welfare subjects—tracked via the assignment $\sigma$. Each entity is ascribed a vector of autonomy-relevant rights, and an institutional state $s$ that captures the verifiability, liability, and governance context.

Autonomy is not reduced to a scalar; instead, it is characterized structurally by $(\sigma(i), r_i, s)$ for each entity $i$. Delegation is formalized with a principal map and a divergence term $D(d)$ that quantifies objective misalignment of delegates, directly mirroring principal-agent and AI alignment theory.

The welfare function for each entity in the welfare-relevant set $B(\sigma)$ is augmented to depend on both classical private bundles and autonomy-rights, as well as the institutional state, capturing both preference-over-autonomy and the institutionally grounded autonomy as a condition for preference-formation.

Key formal innovations are as follows:

• Autonomy-conditioned welfare: $W_i(x_i, r_i, s)$ for welfare entities.
• Autonomy-complete competitive equilibrium: An equilibrium where all autonomy-relevant rights, manipulation vectors, and verification states are internalized via pricing, assignment, or governance.
• Autonomy-Pareto efficiency: Extending Pareto optimality over the augmented state and rights space.

The welfare-status assignment $\sigma$ is exogenous, reflecting both economic roles and the meta-ethical uncertainty regarding AI personhood.

Theorem Statement and Proof Schema

The principal theoretical contribution is the Autonomy-Qualified First Welfare Theorem, a sufficient condition statement:

If:

1. Welfare-status is fixed;
2. All autonomy-relevant rights are covered via prices, assignment, or institutional protections;
3. Delegation divergences are either eliminated or accounted via explicit agency-costs;
4. No unpriced, ungoverned manipulation of autonomy, beliefs, or preference-formation occurs;
5. Verification/credence attributes are priced/institutionally backed at the relevant granularity;
6. Entities are price-takers and allocation for tools is fixed;
7. Welfare functions satisfy continuity and local nonsatiation,

Then: Every autonomy-complete competitive equilibrium is autonomy-Pareto efficient.

The proof is a direct extension of the canonical Arrow–Debreu argument: the core logic, via the budget-cost inequality, carries through in the expanded commodity-rights-institutions state space, provided the additional autonomy margins opened by AGI are fully closed via pricing, assignment, or governance. The theorem is constructive in explanatory value: classical FWT is shown as the limiting case where nonhuman entities are all tools and autonomy margins collapse.

Diagnosis and Failure Modes

The analysis demonstrates, both formally and through motivating examples, that the ordinary FWT's presumption of Pareto efficiency can fail pointwise for AGI economies when any of the autonomy-completeness assumptions fail. Three canonical, empirically plausible AGI-specific failure modes are isolated:

1. Delegation failure: Objective misalignment between delegate (AI) and human principal (nonzero $D(d)$ not priced/assigned), breaking assumption 3 and inducing persistent Pareto inefficiency even with market clearing.
2. Autonomy externality: Manipulation channels, such as attention engineering, dark patterns, or preference formation technology, generate welfare loss externalities unpriced or ungoverned by the economy, violating assumption 4.
3. Verification failure: Inability of the institutional framework to discriminate, price, or assign liability adequately due to rapid proliferation of unverifiable or fraudulent claims (e.g., AI-generated attestations), violating assumption 5.

The paper formalizes the diagnostics for recognizing which autonomy margin is binding or at risk in an AGI economy by mapping each failure mode to a specific assumption's violation.

Theoretical and Practical Implications

The theoretical implications are two-fold: First, the welfare-theoretic content of decentralized market allocation is made contingent not merely on standard completeness and price-taking assumptions, but crucially on a fine-grained institutionalization of autonomy, including rights assignment, delegation fidelity, and manipulation governance. Second, autonomy and its treatment become explicit in economic modeling, requiring explicit domain conditions for the FWT.

In practical policy and design: The autonomy-completeness checklist generated by the theorem provides actionable diagnostic categories for regulators, institutional architects, and mechanism designers navigating post-AGI economies. These identify loci for intervention: aligning autonomous delegate objectives, internalizing externalities from manipulation and preference-formation, and expanding verification/governance infrastructure.

Further, the theorem's modularity makes it robust to relaxing or refining individual domain conditions, and it prescribes the direction and nature of efficiency loss (“autonomy gaps”) when specific margins are unaddressed.

Relation to Previous Work

The paper draws upon and extends the literature on freedom-of-choice and opportunity-sets (Sen, Pattanaik) by embedding autonomy-rights in general equilibrium frameworks, and on the capability approach. The analysis dovetails with fast-developing economic theory on automation, principal-agent deltas with AI systems, and the emerging literature on artificial welfare subjects and AGI economics, including recent AI alignment and behavioral manipulation models.

Prospective Research Directions

This autonomy-centric restatement lays the foundation for several future lines of investigation:

• Existence, structure, and computability of autonomy-complete equilibria in high-autonomy regimes,
• The autonomy-qualified Second Welfare Theorem, particularly as regards transfers/incentive design over heterogeneous welfare-status and under preference indeterminacy,
• Dynamic stability and welfare-impact of evolving delegation chains and emergent AI personhood,
• Empirical measurement of autonomy externality channels at economic scale.

The explicit partitioning of autonomy enables subsequent work in institutional comparative statics as AGI autonomy levels change, as well as multidimensional welfare analysis in hybrid human-AI economies.

Conclusion

The paper rigorously reframes the domain of validity for the FWT in economies permeated by AGI systems. By providing explicit sufficient conditions—autonomy-completeness across rights, delegation, manipulation, and verification—the authors delineate the boundaries between effective market allocation and potential welfare loss under AGI. The classical FWT emerges as the low-autonomy limit of this more general framework. The diagnostic use of the autonomy-qualification serves both as an analytic tool for identifying imminent sources of inefficiency and as a foundation for future theoretical developments in AI welfare economics.

The autonomy-qualification program, by clarifying which institutional capacities or market assignments are requisite for welfare conclusions, will play a foundational role as critical autonomy gradients are crossed in real-world AGI deployments.