The Coasean Singularity: How AI Agents Will Transform Digital Markets

Posted on Oct 24, 2025

Executive Summary

In 2025, researchers from NBER published groundbreaking analysis on how autonomous AI agents will fundamentally reshape digital markets through dramatic reductions in transaction costs. The paper “The Coasean Singularity? Demand, Supply, and Market Design with AI Agents” examines what happens when AI systems can search, negotiate, and transact independently on behalf of humans—potentially reaching a “singularity” where transaction costs approach zero.

Key Insight: AI agents aren’t just tools—they’re market participants that could enable entirely new economic structures previously impossible due to high transaction costs.

The Core Argument: Transaction Costs Drive Everything

The research builds on Ronald Coase’s 1937 insight that transaction costs determine how we organize economic activity. When it’s expensive to search, negotiate, and enforce contracts, we create firms and hierarchies. When those costs drop dramatically, markets can operate in entirely new ways.

The Coasean Singularity Hypothesis: As AI agents reduce transaction costs toward zero, we approach a singularity point where:

  • Preferences can be elicited instantly and costlessly
  • Contracts can be negotiated and enforced automatically
  • Search costs become negligible
  • Identity verification happens seamlessly

This fundamentally changes what kinds of markets and economic structures are feasible.

The Research Framework: Demand, Supply, and Market Design

Demand Side: Why Users Adopt AI Agents

The authors frame user adoption as “derived demand”—users don’t want AI agents per se, they want better outcomes with less effort.

The Trade-Off:

  • Decision Quality: How good is the outcome the agent achieves?
  • Effort Reduction: How much cognitive load does it remove?

Users will adopt agents when the combination of quality and convenience exceeds doing tasks manually or hiring human intermediaries.

Examples of Agent Tasks:

  • Price comparison shopping across thousands of vendors
  • Contract negotiation with optimized terms
  • Job interview scheduling and initial screening
  • Insurance policy comparison with personalized risk analysis
  • Travel planning with real-time optimization

Critical Finding: Adoption depends on task context. Complex, high-stakes decisions (like medical treatment) may see slower adoption than routine transactions (like booking flights).

Supply Side: How Firms Deploy AI Agents

Firms face strategic decisions about agent design, integration, and monetization.

Key Strategic Questions:

  1. Within vs. Across Ecosystem Boundaries

    • Walled Gardens: Agents that only work within one platform (Amazon’s agent only shops Amazon)
    • Open Ecosystems: Agents that operate across multiple platforms (agent compares Amazon, Walmart, Target)

  2. Monetization Models

    • Subscription fees for agent access
    • Commission on transactions facilitated
    • Data monetization from agent interactions
    • Platform fees charged to merchants

  3. Agent Capabilities

    • Basic: Execute pre-defined searches
    • Intermediate: Negotiate within parameters
    • Advanced: Make autonomous decisions with learned preferences

Competitive Dynamics: Firms that create “open” agents may capture more users, but “closed” agents enable better monetization and lock-in.

Market-Level Effects: Efficiency vs. Friction

Efficiency Gains:

  • Reduced Search Costs: Agents can evaluate millions of options instantly
  • Better Matching: Preferences and offerings aligned with precision
  • Lower Communication Costs: Agents negotiate without human involvement
  • Streamlined Contracting: Standard terms executed automatically

New Frictions:

  • Congestion: Millions of agents making requests simultaneously
  • Price Obfuscation: Agents negotiating different prices makes market transparency harder
  • Collusion Risk: Algorithms coordinating on pricing
  • Quality Degradation: “Good enough” automated decisions vs. careful human choices

Net Welfare Effect: Empirically uncertain—depends on how these trade-offs resolve in practice.

Novel Market Designs Enabled by AI Agents

The most exciting implication: AI agents make previously infeasible market mechanisms suddenly viable.

Preference Elicitation at Scale

Traditional Problem: Asking users detailed preferences is costly and annoying.

Agent Solution: Agents can:

  • Observe behavior over time to infer preferences
  • Ask clarifying questions only when necessary
  • Update preference models continuously
  • Handle multi-dimensional trade-offs automatically

New Markets Enabled:

  • Personalized Everything: Every product/service customized to inferred preferences
  • Dynamic Bundling: Optimal combinations updated in real-time
  • Combinatorial Auctions: Complex multi-item bidding previously too expensive to coordinate

Contract Enforcement Without Courts

Traditional Problem: Contract disputes require expensive legal proceedings.

Agent Solution:

  • Smart contracts with automated execution
  • Programmatic dispute resolution
  • Reputation systems with verified transaction history
  • Escrow and verification built into every transaction

New Markets Enabled:

  • Micro-contracts: Agreements too small to justify legal costs become viable
  • Cross-border Trade: Automated enforcement reduces jurisdictional barriers
  • Gig Economy Evolution: Instant matching with enforced quality standards

Identity Verification and Trust

Traditional Problem: Verifying identity and trustworthiness is expensive.

Agent Solution:

  • Continuous identity verification through behavior patterns
  • Cryptographic proof of credentials
  • Cross-platform reputation portability
  • Fraud detection at transaction speed

New Markets Enabled:

  • Zero-trust Marketplaces: Trade with strangers becomes as safe as friends
  • Credential Markets: Skills and credentials verified and portable
  • Reputation Economies: Trust becomes quantifiable and transferable

Regulatory Challenges: When Agents Transact

The paper highlights critical policy questions:

1. Liability and Responsibility

Question: When an AI agent makes a bad decision, who’s responsible?

  • The user who deployed it?
  • The company that built it?
  • The platform that hosts it?

Current Legal Gap: Existing frameworks assume human decision-makers.

2. Algorithmic Collusion

Question: If agents learn to coordinate pricing without explicit instructions, is that illegal?

Challenge: Traditional antitrust law requires “agreement” between humans. Algorithms may coordinate through learned behavior without human conspiracy.

3. Discriminatory Outcomes

Question: If agents learn discriminatory patterns from data, who’s liable?

Complexity: Agents may discover correlations (like zip codes predicting creditworthiness) that have discriminatory effects without explicit bias programming.

4. Market Manipulation

Question: Can agents be designed to manipulate other agents?

Risk: Adversarial agents could exploit vulnerabilities in other agents’ decision-making algorithms.

Practical Implications for Developers and Businesses

For AI Engineers

Design Principles:

  1. Transparency: Make agent decisions auditable
  2. Controllability: Allow users to override or constrain agents
  3. Interoperability: Design for cross-platform operation
  4. Security: Protect against adversarial manipulation

Technical Challenges:

  • Preference learning with limited data
  • Multi-agent coordination without collusion
  • Real-time negotiation at scale
  • Fraud detection in agent-to-agent transactions

For Business Strategists

Strategic Questions:

  1. Build vs. Buy: Develop proprietary agents or license third-party?
  2. Open vs. Closed: Allow agents to operate across platforms or keep them captive?
  3. Monetization: Charge users, merchants, or both?
  4. Differentiation: Compete on agent capability or ecosystem access?

First-mover Advantages:

  • Data network effects (more agent interactions = better models)
  • User habituation (default agent gets preference)
  • Platform entrenchment (merchants integrate with dominant agents)

For Policymakers

Regulatory Priorities:

  1. Clarify Liability: Establish clear rules for agent-caused harms
  2. Enable Competition: Prevent platform lock-in through interoperability requirements
  3. Protect Consumers: Ensure agent transparency and user control
  4. Prevent Collusion: Update antitrust frameworks for algorithmic coordination

Market Design Opportunities:

  • Public agent infrastructure (like digital public goods)
  • Standardized agent protocols (like HTTP for AI agents)
  • Certification systems for agent quality and safety

Research Gaps and Future Work

The authors identify this as “a unique opportunity for economic research to inform real-world policy and market design” because we’re at the very beginning of the AI agent era.

Critical Empirical Questions:

  1. Adoption Curves: How quickly will users trust AI agents with high-stakes decisions?
  2. Quality-Effort Trade-offs: Where do users tolerate lower quality for higher convenience?
  3. Market Structure: Will agent markets concentrate or remain competitive?
  4. Welfare Distribution: Who captures the surplus from reduced transaction costs?

Methodological Challenges:

  • Measuring transaction costs directly is difficult
  • Agent capabilities are evolving rapidly
  • Counterfactual analysis (what would have happened without agents) is complex

Connection to Current AI Developments (October 2025)

This research is remarkably timely given recent developments:

Anthropic’s Computer Use API (October 2024): Claude can now control computers directly, searching, browsing, and transacting—exactly the kind of agent behavior this paper analyzes.

OpenAI’s GPT Store and Custom GPTs: Agents specialized for specific tasks with access to tools and APIs.

Google’s Gemini Agents: Multi-modal agents that can interact with Google’s ecosystem.

The Timing: We’re witnessing the transition from “AI assistants that help” to “AI agents that act independently”—the exact phenomenon this research examines.

Critical Analysis: What the Paper Gets Right and What’s Missing

What’s Compelling

Strong Framework: The demand-supply-market design structure is clean and actionable.

Coase Connection: Building on transaction cost economics grounds the analysis in solid economic theory.

Policy Relevance: Identifying regulatory gaps before they become crises is valuable.

What’s Underdeveloped

Empirical Validation: The paper is primarily theoretical—we need data on actual agent adoption and welfare effects.

Power Dynamics: Limited discussion of how agents might shift power between consumers and firms (or concentrate it further).

Unintended Consequences: More attention to second-order effects would strengthen the analysis (e.g., unemployment from agent intermediation).

Technical Feasibility: The paper assumes agents will become highly capable—what if technical barriers persist?

Implications for AI Practitioners

For Engineers Building AI Agents

Key Takeaways:

  1. Design for Interoperability: Closed ecosystems may not win long-term
  2. Prioritize Explainability: Users need to understand and trust agent decisions
  3. Build in Constraints: Allow users to set boundaries on agent behavior
  4. Plan for Scale: Agent-to-agent interactions will create congestion challenges

Technical Priorities:

  • Preference learning with privacy preservation
  • Multi-agent coordination protocols
  • Adversarial robustness against manipulation
  • Real-time decision-making at transaction speed

For Companies Deploying Agents

Strategic Considerations:

  1. Where do transaction costs hurt most in your industry? (Start there with agents)
  2. Can you create network effects with agent data? (Defensibility)
  3. Will you face regulatory scrutiny for agent decisions? (Plan for compliance)
  4. How will competitors respond to your agent strategy? (Game theory)

Monetization Models to Test:

  • Freemium (basic agent free, premium features paid)
  • Transaction fees (commission on agent-facilitated deals)
  • Subscription (monthly fee for agent access)
  • Data licensing (aggregate insights from agent behavior)

The “Singularity” Question: Will Transaction Costs Really Approach Zero?

The paper’s title asks whether we’ll reach a “Coasean Singularity”—but doesn’t definitively answer.

Arguments For:

  • AI capabilities are improving exponentially
  • Marginal cost of agent operations approaches zero
  • Network effects accelerate adoption
  • Competition drives agents toward perfect efficiency

Arguments Against:

  • Human trust takes time to build
  • Regulatory friction creates artificial transaction costs
  • Adversarial dynamics (manipulation, fraud) persist
  • Coordination problems among agents don’t disappear

Most Likely Outcome: Asymptotic approach—transaction costs drop dramatically but never quite reach zero due to persistent frictions (regulation, trust, coordination challenges).

Conclusion: A Framework for the Agent Economy

“The Coasean Singularity?” provides essential scaffolding for thinking about how AI agents will reshape markets. By framing the question through demand, supply, and market design, the authors give us tools to analyze specific agent deployments and predict their economic effects.

The Big Picture: We’re moving from:

  • Human-mediated markets (high transaction costs, simple mechanisms)
  • To agent-mediated markets (low transaction costs, complex mechanisms possible)

This transition will:

  • Enable new market designs previously infeasible
  • Create regulatory challenges we’re unprepared for
  • Shift power dynamics between consumers, firms, and platforms
  • Generate efficiency gains (but with uncertain distribution)

For Practitioners: The time to think about agent strategy is now. The companies building agent infrastructure today will shape how these markets evolve.

For Researchers: Empirical work is desperately needed. We need data on:

  • Actual adoption patterns across contexts
  • Measured welfare effects (not just theoretical)
  • Platform competition dynamics
  • Regulatory effectiveness

For Policymakers: Proactive regulation is essential. Waiting for problems to emerge means playing catch-up in fast-moving markets.

The Coasean Singularity may not arrive tomorrow—but the forces driving toward it are already in motion. Understanding the economics of AI agents isn’t optional for anyone working at the intersection of technology, markets, and policy.

References and Further Reading

Primary Source:

  • Shahidi, P., Rusak, G., Manning, B.S., Fradkin, A., & Horton, J.J. (2025). “The Coasean Singularity? Demand, Supply, and Market Design with AI Agents.” The Economics of Transformative AI, Chapter 6. University of Chicago Press. NBER. Full PDF

Related NBER Research:

  • The Economics of Artificial Intelligence: An Agenda (2019)
  • The Economics of Transformative AI (2025)

Foundational Economics:

  • Coase, R.H. (1937). “The Nature of the Firm.” Economica, 4(16), 386-405.
  • Williamson, O.E. (1979). “Transaction-Cost Economics: The Governance of Contractual Relations.” Journal of Law and Economics, 22(2), 233-261.

Contemporary AI Agent Development:

  • Anthropic Computer Use API Documentation
  • OpenAI Custom GPTs and GPT Store
  • Google Gemini Agents

Author Note: This summary was created to make cutting-edge economic research on AI agents accessible to practitioners. The original paper is highly technical—this summary extracts actionable insights for engineers, business strategists, and policymakers working on AI agent systems.

Published: October 24, 2025 Last Updated: October 24, 2025 Reading Time: 15 minutes Technical Level: Intermediate to Advanced