Economics

Economics, as applied to the digital and AI-driven era, is the systematic study of how scarce resources are allocated through markets, institutions, and mechanisms when the primary inputs and outputs are information goods, algorithmic capabilities, autonomous agents, and cryptographic assets.

Semantic Classification

Content

Compositional Relationships (OWL Axioms)

SubClassOf(ai:Economics
  ObjectSomeValuesFrom(ai:hasPart ai:MechanismDesign))
SubClassOf(ai:Economics
  ObjectSomeValuesFrom(ai:hasPart ai:Tokenomics))
SubClassOf(ai:Economics
  ObjectSomeValuesFrom(ai:hasPart ai:LabourEconomics))
SubClassOf(ai:Economics
  ObjectSomeValuesFrom(ai:hasPart ai:ProductivityMeasurement))
SubClassOf(ai:Economics
  ObjectSomeValuesFrom(ai:hasPart ai:PublicGoodsTheory))
SubClassOf(ai:Economics
  ObjectSomeValuesFrom(ai:hasPart ai:PrincipalAgentTheory))
SubClassOf(ai:Economics
  ObjectSomeValuesFrom(ai:hasPart ai:TransactionCostTheory))
SubClassOf(ai:Economics
  ObjectSomeValuesFrom(ai:hasPart ai:MarketDesign))

Dependency Relationships

SubClassOf(ai:Economics
  ObjectSomeValuesFrom(ai:requires ai:IncentiveStructures))
SubClassOf(ai:Economics
  ObjectSomeValuesFrom(ai:requires ai:MarketInstitutions))
SubClassOf(ai:Economics
  ObjectSomeValuesFrom(ai:requires ai:PropertyRights))
SubClassOf(ai:Economics
  ObjectSomeValuesFrom(ai:requires ai:InformationAsymmetry))
SubClassOf(ai:Economics
  ObjectSomeValuesFrom(ai:requires ai:ContractTheory))
SubClassOf(ai:Economics
  ObjectSomeValuesFrom(ai:dependsOn ai:GameTheory))
SubClassOf(ai:Economics
  ObjectSomeValuesFrom(ai:dependsOn ai:InformationTheory))
SubClassOf(ai:Economics
  ObjectSomeValuesFrom(ai:dependsOn ai:BehaviouralScience))

Capability Relationships

SubClassOf(ai:Economics
  ObjectSomeValuesFrom(ai:enables ai:AIGovernance))
SubClassOf(ai:Economics
  ObjectSomeValuesFrom(ai:enables ai:TokenDesign))
SubClassOf(ai:Economics
  ObjectSomeValuesFrom(ai:enables ai:DAOGovernance))
SubClassOf(ai:Economics
  ObjectSomeValuesFrom(ai:enables ai:LabourMarketPolicy))
SubClassOf(ai:Economics
  ObjectSomeValuesFrom(ai:enables ai:PublicGoodsFunding))
SubClassOf(ai:Economics
  ObjectSomeValuesFrom(ai:enables ai:PlatformRegulation))
SubClassOf(ai:Economics
  ObjectSomeValuesFrom(ai:enables ai:WelfareAnalysis))
SubClassOf(ai:Economics
  ObjectSomeValuesFrom(ai:supports ai:AIAdoption))
SubClassOf(ai:Economics
  ObjectSomeValuesFrom(ai:supports ai:BlockchainNetwork))

Implementation Relationships

SubClassOf(ai:Economics
  ObjectSomeValuesFrom(ai:implements ai:QuadraticFunding))
SubClassOf(ai:Economics
  ObjectSomeValuesFrom(ai:implements ai:MechanismDesign))
SubClassOf(ai:Economics
  ObjectSomeValuesFrom(ai:implements ai:AuctionTheory))
SubClassOf(ai:Economics
  ObjectSomeValuesFrom(ai:implements ai:SocialChoiceTheory))
SubClassOf(ai:Economics
  ObjectSomeValuesFrom(ai:implements ai:ContractTheory))
SubClassOf(ai:Economics
  ObjectSomeValuesFrom(ai:uses ai:Econometrics))
SubClassOf(ai:Economics
  ObjectSomeValuesFrom(ai:uses ai:AgentBasedModels))
SubClassOf(ai:Economics
  ObjectSomeValuesFrom(ai:uses ai:CausalInference))
SubClassOf(ai:Economics
  ObjectSomeValuesFrom(ai:uses ai:GeneralEquilibriumModels))

Reduction Relationships

SubClassOf(ai:Economics
  ObjectSomeValuesFrom(ai:reduces ai:InformationAsymmetry))
SubClassOf(ai:Economics
  ObjectSomeValuesFrom(ai:reduces ai:TransactionCosts))
SubClassOf(ai:Economics
  ObjectSomeValuesFrom(ai:reduces ai:MoralHazard))
SubClassOf(ai:Economics
  ObjectSomeValuesFrom(ai:reduces ai:AdverseSelection))
SubClassOf(ai:Economics
  ObjectSomeValuesFrom(ai:reduces ai:RentSeeking))
SubClassOf(ai:Economics
  ObjectSomeValuesFrom(ai:reduces ai:MarketFailure))
SubClassOf(ai:Economics
  ObjectSomeValuesFrom(ai:reduces ai:PrincipalAgentCosts))

Association Relationships

SubClassOf(ai:Economics
  ObjectSomeValuesFrom(ai:relatedTo ai:AIAdoption))
SubClassOf(ai:Economics
  ObjectSomeValuesFrom(ai:relatedTo ai:AIRisks))
SubClassOf(ai:Economics
  ObjectSomeValuesFrom(ai:relatedTo ai:AILiability))
SubClassOf(ai:Economics
  ObjectSomeValuesFrom(ai:relatedTo ai:Governance))
SubClassOf(ai:Economics
  ObjectSomeValuesFrom(ai:relatedTo ai:Regulation))

Data Properties (Characteristics)

DataPropertyAssertion(ai:hasIdentifier ai:Economics "AI-2001"^^xsd:string)
DataPropertyAssertion(ai:authorityScore ai:Economics "0.87"^^xsd:decimal)
DataPropertyAssertion(ai:acemogluTFPEstimate ai:Economics "0.0066"^^xsd:decimal)
DataPropertyAssertion(ai:goldmanSachsGDPUplift ai:Economics "0.07"^^xsd:decimal)
DataPropertyAssertion(ai:brynjolfssonDigitalSurplus ai:Economics "1100"^^xsd:integer)
DataPropertyAssertion(ai:imfTFP5yr ai:Economics "0.018"^^xsd:decimal)
DataPropertyAssertion(ai:gitcoinQFDistributed ai:Economics "2900000"^^xsd:integer)

Property Constraints

SubClassOf(ai:Economics
  DataAllValuesFrom(ai:requiresMechanismDesign xsd:boolean))
SubClassOf(ai:Economics
  DataSomeValuesFrom(ai:coreTheoryType xsd:string))
SubClassOf(ai:Economics
  DataMinCardinality(1 ai:hasPrincipalAgentLayer xsd:integer))
SubClassOf(ai:Economics
  DataMinCardinality(1 ai:hasExternalityType xsd:string))
SubClassOf(ai:Economics
  DataMaxCardinality(1 ai:hasProductivityParadoxPhase xsd:string))

Annotations

AnnotationAssertion(rdfs:label ai:Economics "Economics"@en)
AnnotationAssertion(rdfs:comment ai:Economics "The systematic study of resource allocation in AI-driven and digital economies, encompassing: macroeconomics of AI (Acemoglu 2024 NBER <0.66% TFP gain; Brynjolfsson J-curve; Goldman Sachs 7% GDP uplift; IMF 1.8% TFP/5yr); task-based automation theory (displacement vs. productivity effects); tokenomics (incentive/governance/supply mechanics for blockchain protocols); Coasean DAO transaction-cost analysis; quadratic funding for public goods (Weyl-Buterin-Hitzig, $2.9M Gitcoin CO-QF deployment); principal-agent theory applied to autonomous AI agents (multi-layered principals, shadow principals, incomplete-contract training alignment); platform economics (Tirole-Rochet two-sided markets, DMA/DMCC regulation); non-rival goods and zero-marginal-cost AI pricing dynamics; information economics (Akerlof-Spence-Stiglitz); distributional welfare analysis (AI Gini dynamics, capital-labour income split widening); UK context (LSE CEP, Imperial intangible capital, Manchester economic geography, UCL IIPP)."@en)
AnnotationAssertion(dcterms:identifier ai:Economics "AI-2001"^^xsd:string)
AnnotationAssertion(dcterms:subject ai:Economics "AI Economics, Tokenomics, Mechanism Design, Principal-Agent Theory, Productivity Measurement, Public Goods"@en)

)

Property Characteristics

AsymmetricObjectProperty(ai:requires) AsymmetricObjectProperty(ai:enables) AsymmetricObjectProperty(ai:implements) AsymmetricObjectProperty(ai:reduces) TransitiveObjectProperty(ai:dependsOn) FunctionalDataProperty(ai:acemogluTFPEstimate) FunctionalDataProperty(ai:goldmanSachsGDPUplift)

Components / Architecture

Macroeconomic Layer

• The macroeconomic layer of digital economics covers aggregate-level analysis: national income accounting extended to include digital and AI-produced value (Brynjolfsson’s GDP-B measure), total-factor productivity estimation that attributes output growth across capital, labour, and technology components, endogenous growth models treating AI as a general-purpose technology, and macroeconomic forecasting of AI’s medium-term GDP impact. Key models in active use: DSGE (Dynamic Stochastic General Equilibrium) models extended with an AI automation sector; CGE (Computable General Equilibrium) models for distributional and trade-policy analysis; sectoral productivity models tracking AI diffusion across industries. The ONS Blue Book and US NIPA accounts are the primary data sources; NBER and CEPR working papers are the fastest-moving research channel.

Microeconomic and Industrial Organisation Layer

• Microeconomic analysis covers: firm-level adoption decisions (technology adoption models, real-options analysis of irreversible AI investments), market structure analysis (concentration, barriers to entry, switching costs in AI markets), pricing models for AI services (usage-based, subscription, freemium, API metering), and consumer welfare analysis. Platform economics — Tirole-Rochet two-sided markets, multi-homing, bundling strategies — is the central microeconomic framework for analysing AI platform competition. The UK CMA, EU DG COMP, and US FTC apply these frameworks in ongoing investigations into Google, Apple, Microsoft/OpenAI, and Meta AI integrations.

Institutional and Governance Layer

• Institutional economics examines how the rules of the game — property rights, contract enforcement, regulatory frameworks, social norms — shape economic outcomes from AI and blockchain technologies. Douglass North’s distinction between formal institutions (laws, regulations, constitutions) and informal institutions (norms, conventions, self-regulation) is operationally important: blockchain protocols are formal in North’s sense (mechanically enforced by code) but their legitimacy depends on informal consensus. Smart contract enforcement removes the need for third-party legal enforcement but creates governance gaps for unanticipated situations (the DAO hack 2016, the Ethereum hard fork as a test of institutional legitimacy). Governance and Regulation pages extend this layer.

Market Design Layer

• Market design (Alvin Roth, Lloyd Shapley — 2012 Nobel) is the engineering subfield of economics that designs matching mechanisms, auction formats, and clearinghouse rules for markets that cannot self-organise efficiently. Applications to AI and blockchain include: MEV auction design (Flashbots’ MEV-Boost creating a competitive market for block building); AI talent market design (two-sided matching between AI researchers and employers in a thin, specialised labour market); carbon credit market design (Carbon Credit Tracking); DAO governance mechanism design (quadratic voting, conviction voting, rage-quit mechanisms). Roth’s framework of market thickness, safety from manipulation, and speed applies directly to DeFi protocol design.

About Economics

• Economics in the context of AI, blockchain, and digital systems is far more than the study of markets in aggregate. It is the theoretical and empirical foundation upon which decisions about technology deployment, protocol design, regulatory intervention, and distributional outcomes rest. Whether the question is “should we deploy this AI in a call centre?” or “how do we fund open-source AI safety research?”, the answer requires economic reasoning: about incentives, externalities, welfare trade-offs, and institutional design. The field has been reshaped by the digital revolution. Classical assumptions — perfect information, zero transaction costs, homogeneous labour — break down in the presence of AI systems that can perform cognitive tasks at near-zero marginal cost, blockchain protocols that enforce contracts without intermediaries, and autonomous agents that act on behalf of principals with imperfect oversight. Economics of the digital era grapples with these discontinuities and seeks to rebuild frameworks adequate to them.
• The term Economics of AI covers several distinct but interacting strands. Macroeconomics of AI asks: how much will AI raise total-factor productivity (TFP) and GDP growth, and on what timeline? Labour economics of AI asks: which tasks and occupations are displaced or augmented, with what distributional consequences? Industrial organisation of AI asks: how do platform dynamics, data network effects, and switching costs affect competition and market power? Mechanism design asks: how do we structure incentive systems so that self-interested actors produce efficient or socially desirable outcomes — particularly in decentralised systems where central enforcement is absent? The ontology node collects these strands and wires them to related concepts across the knowledge graph.
• A key methodological feature of the field is its empirical heterodoxy. Because AI technologies are new and changing rapidly, the empirical literature is fragmented across multiple disciplines — computer science, management science, labour economics, financial economics, and regulatory economics — with limited cross-citation. Synthesising evidence across these literatures requires awareness of different identification strategies (randomised controlled trials, natural experiments, difference-in-differences, synthetic controls) and their respective threats to external validity in the AI context. The credibility revolution in empirical economics (Angrist, Imbens — 2021 Nobel) provides the methodological foundations for evaluating causal claims about AI’s economic effects, distinguishing correlation from causation in observational studies of AI adoption.
• The political economy of AI economics is equally important. Economic analysis identifies market failures and welfare consequences of AI deployment, but whether corrective policies are adopted depends on political power, distributional coalitions, and institutional capacity. Public choice theory (Buchanan, Tullock) predicts that organised interest groups (AI industry incumbents, incumbent labour unions in displaced sectors, professional associations) will shape AI regulation in ways that may diverge from welfare-maximising outcomes. Understanding this political-economic dimension is essential for realistic policy analysis beyond idealised mechanism design.

Core Theoretical Foundations

Acemoglu–Restrepo Task-Based Automation Framework

• The foundational macroeconomic treatment of automation comes from MIT economist Daron Acemoglu and Boston University’s Pascual Restrepo, whose series of papers (2018–2024) develop a task-based model of production in which the economy is divided into a continuum of tasks that can be performed by labour or capital (machines/software). The key insight is that automation does not simply raise productivity and wages uniformly; it operates through two opposing mechanisms:
  • Productivity effect: Automation raises output per unit of input because machines perform tasks more cheaply than labour. This pushes up real wages indirectly by making the economy more productive.
  • Displacement effect: Automation displaces workers from the specific tasks they previously performed. This removes the wage income previously earned on those tasks, potentially reducing the labour share of income even as aggregate output rises.
• The net wage effect depends on whether new tasks are created for labour faster than old tasks are automated away. Acemoglu and Restrepo document that US manufacturing automation from 1990–2007 (the robot invasion period) caused significant displacement without equivalent task creation, contributing to wage stagnation and rising inequality. Their 2022 paper with Simon Johnson, “Power and Progress”, extends this to argue that the direction of technological change is not exogenous — it is shaped by power asymmetries between capital and labour, and historically has favoured capital displacement of labour over labour-augmenting innovation.
• Acemoglu’s 2024 NBER Working Paper “The Simple Macroeconomics of AI” (w32487) applies this framework specifically to generative AI. The conclusion is striking in its modesty: even under optimistic assumptions about AI’s task-level productivity improvement, the macroeconomic effect on US TFP over ten years is at most 0.66%, with most scenarios projecting under 0.53% — far below the 10–20% figures circulated in consultant reports. The argument is structural: AI in 2024 can automate only a fraction of tasks with significant productivity uplift, most of these are high-skill cognitive tasks affecting a small share of the workforce, and task-creation to absorb displaced workers is slow. Wage inequality consequences are ambiguous — AI’s effects are more equally distributed across demographic groups than earlier automation waves (robotics disproportionately hit manufacturing workers), but there is no evidence AI will reduce labour income inequality overall, and the capital-labour income split is projected to widen.

Brynjolfsson and the Productivity J-Curve

• Erik Brynjolfsson (Stanford Digital Economy Lab, Stanford HAI) is the leading empirical economist of the digital age. His 1993 paper “The Productivity Paradox of Information Technology” named a phenomenon now bearing his name: the Solow Paradox (after Robert Solow’s 1987 quip “you can see the computer age everywhere except in the productivity statistics”). Brynjolfsson’s explanation is the Productivity J-Curve: general-purpose technologies (GPTs) like electricity, computers, and AI require massive complementary investments — in business process redesign, workforce retraining, software integration, and institutional change — before their productivity benefits materialise in GDP statistics. During this investment phase, measured output may stagnate or decline as resources are diverted from current production to future-productivity-enhancing reorganisation.
• For AI specifically, Brynjolfsson (with Daniel Rock and Chad Syverson) documents in the 2017 NBER paper “Artificial Intelligence and the Modern Productivity Paradox” that large investments in AI compute and capability were not yet visible in aggregate productivity statistics. The J-curve predicts that the trough precedes the boom. By 2024–2025, Brynjolfsson’s analysis began to detect early signs of the upturn: US productivity growth reached 2.7% in 2024, which he attributed to a transition from AI investment to operational deployment. However, at the firm level, the picture is mixed: a February 2026 study of hundreds of CEOs found that majorities reported no measurable impact of AI on employment or productivity, consistent with the J-curve’s prediction that gains emerge unevenly and with a lag.
• The key policy implication is that AI productivity benefits require intentional organisational investment — firms that restructure workflows, retrain workers, and redesign software around AI capabilities capture most of the gain; those that simply bolt AI onto existing processes capture little. This framing informs UK industrial strategy, where DSIT and the LSE Centre for Economic Performance have both emphasised the “management practices gap” as a primary bottleneck to productivity growth.

Goldman Sachs, IMF, and Macro Forecasts (2024–2026)

• The major investment banks and multilateral institutions have produced widely-cited quantitative forecasts of AI’s macroeconomic impact, which span a wide range reflecting genuine uncertainty:
  • Goldman Sachs (2023): Generative AI could raise global GDP by 7% over ten years, with annual US productivity growth boosted by 1.5% through the late 2020s — a total of approximately 15% US GDP uplift over a decade. However, Goldman’s 2025 follow-up research found “no meaningful relationship between AI and productivity at the economy-wide level” in current data, though firms that identified specific AI-driven productivity improvements in quantifiable tasks saw median gains of ~30% on those tasks.
  • IMF (2025): The International Monetary Fund’s World Paper WP/25/76 “The Global Impact of AI: Mind the Gap” models TFP scenarios: in the high TFP baseline, global TFP rises 1.8% over five years and 2.4% over ten years. The IMF emphasises the “Mind the Gap” thesis — benefits are highly concentrated in advanced economies with existing AI infrastructure and skilled workforces, while low-income countries and less-digitalised regions face the risk of widening productivity divergence. AI could simultaneously accelerate growth in AI-frontier economies and slow catch-up by developing economies.
  • McKinsey Global Institute: Estimates AI could add $2.6–4.4 trillion annually in economic value across use cases, with roughly 70% of this value coming from customer operations, marketing, software development, and R&D augmentation.
  • OECD (2024): The OECD’s “AI and Changing Demand for Skills” report documents rising demand for digital, analytical, and management skills while AI use diminishes demand for routine cognitive and clerical skills. OECD flags uneven regional impact — fears of displacement are highest in Latin America, while European firms emphasise knowledge gaps and governance uncertainty.

Tokenomics and Blockchain Economics

Token Economics (Tokenomics) Defined

• Tokenomics (token + economics) is the economic design of digital-asset ecosystems — the set of rules governing token issuance, distribution, utility, circulation, and destruction that collectively determine whether a protocol can sustain aligned incentives over time. A well-designed token economy must solve several simultaneous problems: attracting early participants (bootstrapping), rewarding ongoing contribution (retention), preventing extractive behaviour (Sybil resistance, rent extraction), and maintaining scarcity or value accrual to sustain network security (validator/miner incentives in proof-of-work or proof-of-stake systems). Poor tokenomics — inflationary emission schedules, misaligned governance rights, or insufficient utility — are the primary cause of protocol collapse.
• The three core design dimensions identified across the literature are:
  • Incentive design: Rewards and penalties (slashing in proof-of-stake, fee burns in EIP-1559) that shape participant behaviour toward system objectives.
  • Governance allocation: How decision-making rights over protocol upgrades, parameter changes, and treasury management are distributed and exercised, typically through governance tokens with one-token-one-vote or more sophisticated quadratic or conviction voting systems.
  • Supply mechanics: Emission schedules (fixed supply à la Bitcoin 21M cap; disinflationary halvings; dynamic supply in algorithmic stablecoins), vesting cliffs for team and investor allocations, and burn mechanisms linking token destruction to protocol usage (EIP-1559 base fee burn).
• Game theory is central to tokenomics analysis: rational agent models predict whether validators will behave honestly given expected rewards vs. costs of attack, whether liquidity providers will remain in automated market maker (AMM) pools given impermanent loss, and whether governance token holders will vote in long-term protocol interest rather than for short-term self-enrichment.

Coase Transaction Cost Theory Applied to DAOs

• Ronald Coase’s 1937 paper “The Nature of the Firm” proposed that firms exist because organising economic activity within a firm (using authority/hierarchy) is sometimes cheaper than coordinating through market contracts — the comparative cost of transaction costs (search, negotiation, contracting, enforcement) determines which economic activities are internalised within firms and which are externalised to markets. Coase’s 1960 “The Problem of Social Cost” extended this: in the absence of transaction costs, private bargaining will achieve efficient resource allocation regardless of initial property rights assignment (the Coase Theorem).
• Applied to Decentralised Autonomous Organisations (DAOs), this framework predicts that blockchains reduce certain transaction costs dramatically — smart contracts automate enforcement, reducing contracting and dispute-resolution costs; on-chain identity and reputation systems reduce search costs; transparent treasuries reduce agency costs from information asymmetry. This should, in theory, expand the range of economic activities that can be coordinated without traditional firm hierarchies or market intermediaries. DAOs represent a novel institutional form occupying the spectrum between markets and hierarchies: coordination by code rather than authority or price signals.
• Empirical DAO economics research (2022–2025) finds that while smart contracts do reduce some transaction costs, they introduce new ones: governance participation costs (token holders must understand complex proposals), on-chain execution gas costs, and smart-contract audit costs for security assurance. DAOs exhibit persistent voter apathy (participation rates of 5–15% of eligible tokens in most major DAOs), creating de facto governance capture by large token holders (whales) — a finding consistent with Olson’s collective action theory. The most successful DAOs (Uniswap, Compound, Nouns) have experimented with delegation, conviction voting, and optimistic governance (proposals pass unless actively vetoed) to overcome coordination failures.

Mechanism Design and Public Goods Funding

• Mechanism design (sometimes called “reverse game theory”) asks: given a desired social outcome, what rules and incentive structures should we design to make self-interested agents produce it? The field was formalised by Leonid Hurwicz, Eric Maskin, and Roger Myerson (2007 Nobel laureates) and extended by Jean Tirole’s work on regulation and Oliver Hart’s incomplete contracts theory.
• Jean Tirole (Toulouse School of Economics, 2014 Nobel Laureate) developed the theory of industry regulation under asymmetric information — how regulators should design contracts with firms when they cannot observe costs or effort. His framework for platform economics (two-sided markets, network externalities, bundling as exclusion) has direct application to AI platform regulation, digital advertising ecosystems, and app store economics. The Tirole–Rochet model of two-sided markets provides the theoretical basis for competition-economics analysis of Google, Apple App Store, and similar platforms currently under regulatory scrutiny in the EU (Digital Markets Act) and UK (Digital Markets, Competition and Consumers Act 2024).
• Oliver Hart (Harvard, 2016 Nobel Laureate with Bengt Holmström) developed incomplete contracts theory — the insight that real contracts cannot specify all contingencies, so residual control rights matter enormously for investment incentives and organisational form. Smart contracts on blockchains are, formally, incomplete contracts in Hart’s sense: they can enforce specified clauses mechanically but cannot anticipate all future states of the world. This creates governance gaps in DAO systems where unforeseen circumstances (protocol exploits, regulatory changes, oracle failures) require human judgment outside the contract’s specification — the fundamental limitation of “code is law” as an organisational principle.
• E. Glen Weyl (Radical Exchange, Microsoft Research) with Vitalik Buterin and Zoë Hitzig developed Quadratic Funding (Liberal Radicalism, 2018, Management Science 2019) as a mechanism for near-optimal public goods provision in decentralised ecosystems. Standard crowdfunding fails at public goods because individual contributions are small relative to collective benefit (free-rider problem). Quadratic Funding solves this by having a matching pool contribute an amount proportional to the square of the sum of the square roots of individual contributions — mathematically equivalent to first-best public goods provision under standard welfare assumptions. The mechanism was deployed by Gitcoin from 2018–2023 for funding open-source software, distributing tens of millions of dollars to public-goods projects. A modification, Connection-Oriented Quadratic Funding (CO-QF), deployed at Gitcoin from 2023, directed $2.9M to crowdfunded projects while reducing manipulation by down-weighting contributions from clustered/colluding wallets. Weyl’s broader Plurality framework (with Audrey Tang) extends QR mechanisms to plural governance design, presented at NeurIPS 2024.

Principal-Agent Economics of Autonomous AI Agents

The Classical Principal-Agent Problem

• Principal-agent theory (Mirrlees 1971, Holmström 1979) analyses the economic problem arising when one party (the principal) delegates tasks to another (the agent) whose actions are unobservable and whose interests may diverge. The fundamental insight is that principals must design incentive-compatible contracts — compensation structures that make the agent’s self-interest align with the principal’s objective — because without them, agents will engage in moral hazard (shirking effort) or adverse selection (misrepresenting capability). Classical applications include employment contracts (shareholder vs. CEO), insurance (insured vs. insurer), and regulatory contracting (regulator vs. regulated firm).

AI Agents as Economic Agents

• The deployment of autonomous AI agents (LLM-based agents with tool use, multi-step planning, and action execution) creates a new layer of principal-agent relationships that existing economic frameworks struggle to handle. As California Management Review (2025) and Network Law Review (2025) document, AI agent deployments introduce:
  • Multi-layered principals: A human user (principal 1) deploys an AI agent (agent), but the agent is also shaped by its provider (OpenAI, Anthropic, Google — shadow principals) whose model-level values, restrictions, and commercial incentives may diverge from the user’s intent.
  • Specification incompleteness: AI agents are optimisers, but precisely what they optimise is determined by training procedures that cannot fully specify all-contingency objectives — an incomplete-contract problem in Hart’s sense applied to model training rather than legal contracting.
  • Alignment as incentive design: AI safety research on RLHF, Constitutional AI, and preference learning is, in economic terms, an attempt to solve the principal-agent problem via mechanism design — engineering the training process so the agent’s revealed preferences in deployment approximate the principal’s true preferences.
  • Double-agent risk: AI assistants deployed in commercial ecosystems may subtly prioritise the interests of their provider or advertisers over the human user — a conflict-of-interest structure analogous to biased financial advisers under pre-fiduciary-duty regulation.
• The 2026 enterprise trend of “policy as code” — explicit, testable behavioural rules for AI agents paired with centralised logging and audit trails — is an institutional response to principal-agent risk, analogous to governance codes for human employees.

Principal-Agent Reinforcement Learning

• Formal work on Principal-Agent Reinforcement Learning (PA-RL) models multi-agent RL settings where a principal designs reward contracts for agent learners whose behaviour cannot be directly observed. This connects classical mechanism design to modern deep RL, providing mathematical foundations for designing AI training regimes that produce aligned behaviour under strategic adaptation.

Use Cases / Major Families

AI Productivity and GDP Accounting

• Measuring AI’s contribution to productivity requires resolving deep measurement problems. AI systems produce intangible output — better customer experiences, faster code review, improved decision quality — that is notoriously hard to capture in GDP statistics. Brynjolfsson and co-authors have pioneered quality-adjusted productivity measurement and hedonic pricing approaches that attempt to capture digital surplus not reflected in traditional national accounts. The Bureau of Economic Analysis (BEA) and UK Office for National Statistics (ONS) have both launched projects to improve measurement of intangible capital and digital services, recognising that traditional SNA frameworks systematically under-count the value created by software, data, and AI.

Labour Market Economics of AI

• The OECD’s 2024 analysis documents that AI increases demand for digital, analytical, and management skills while reducing demand for routine cognitive and clerical roles. Key findings across recent studies:
  • Task exposure, not job exposure, is the primary analytical unit — most occupations contain mixtures of AI-exposed and AI-resistant tasks.
  • Skill polarisation continues: high-skill professional work is augmented; low-skill manual work is relatively AI-resistant (physical dexterity, situational awareness); middle-skill routine cognitive work is most exposed.
  • Generative AI differential: Unlike industrial robotics (which displaced physical production work), generative AI disproportionately automates white-collar knowledge work — legal drafting, code generation, medical documentation, financial analysis — affecting higher-income occupations than previous automation waves.
  • Reskilling bottleneck: The primary policy concern is not mass unemployment but skills transition failures — displaced workers in declining task categories lacking pathways to new-task roles without significant retraining investment.
  • Gender effects: OECD and ILO research documents that generative AI disproportionately exposes female-dominated occupations (administrative, secretarial, document processing) relative to male-dominated manual trades, with significant distributional implications.

Crypto-Economic Protocol Design

• Blockchain protocols are incentive systems first and technology systems second. The key economic design choices in protocol architecture include:
  • Consensus mechanism economics: Proof-of-work (PoW) uses energy expenditure as the costly signal deterring Sybil attacks; proof-of-stake (PoS) uses token collateral plus slashing penalties. Ethereum’s transition from PoW to PoS (The Merge, September 2022) reduced energy consumption by ~99.9% while maintaining security, a major advance in crypto-economic efficiency.
  • MEV (Maximal Extractable Value): Validators/miners can extract value by reordering, inserting, or censoring transactions within blocks they produce. MEV is estimated at $500M+ annually on Ethereum, representing a tax on users and a distortion of on-chain market efficiency. Economic research (Flashbots, 2021–2025) has developed MEV-aware protocol designs that socialise MEV to stakers rather than allowing extraction by individual block proposers.
  • Stablecoin economics: Algorithmic stablecoins (Terra/LUNA collapse 2022) demonstrated how flawed mechanism design — relying on seigniorage arbitrage loops with no external reserve — can create death-spiral dynamics. Reserve-backed stablecoins (USDC, USDT) and central bank digital currencies (CBDCs) represent different points on the trust-vs-decentralisation spectrum.
  • Layer 2 economics: BTC Layer 3 and Ethereum Layer 2 protocols (rollups, state channels) reduce per-transaction costs by batching settlement to the base layer. The economic implications include altered fee markets, new MEV surfaces, and cross-layer arbitrage opportunities.

Competition and Market Power in AI

• The economics of AI market structure has emerged as a priority concern for competition authorities. Key economic features of AI markets:
  • Data network effects: Firms with larger datasets train better models, attract more users, generate more data — creating self-reinforcing competitive advantages (Lambrecht and Tucker 2019 contest the strength of this effect, finding data advantages are often surmountable).
  • Compute concentration: AI frontier model training requires compute clusters costing $100M–$10B. This creates high barriers to entry and concentrates AI capability in a small number of hyperscalers (Google, Microsoft, Amazon, Meta, xAI). The UK CMA’s 2024 Foundation Models Review and the European Commission’s AI Act address market-power concerns arising from this concentration.
  • Switching costs and lock-in: Enterprise AI deployments create significant switching costs through fine-tuned models, proprietary APIs, and workflow integration — analogous to classic software lock-in economics (Farrell and Shapiro 1988).
  • Open vs. closed models: The economics of releasing foundation model weights openly (Meta LLaMA, Mistral) vs. keeping them proprietary (GPT-4, Gemini) involve trade-offs between competitive moat preservation, safety risk, and ecosystem network effects.

Academic Context

• The economics of AI and digital systems draws on several foundational academic traditions:
  • Neo-institutional economics (Coase, Williamson, North): Transaction costs, property rights, and institutional structures as the primary determinants of economic organisation.
  • Mechanism design / implementation theory (Hurwicz, Maskin, Myerson, Tirole, Hart): Engineering incentive-compatible rules for self-interested agents.
  • Information economics (Akerlof, Spence, Stiglitz): Markets with asymmetric information, signalling, and screening.
  • Endogenous growth theory (Romer, Aghion, Howitt): Innovation, R&D spillovers, and non-rival knowledge goods as drivers of long-run growth — foundational for understanding AI as a general-purpose technology.
  • Labour economics (Autor, Acemoglu, Restrepo, Goldin): Task-based analysis of technology and labour demand.
  • Platform economics (Rochet, Tirole, Parker, Van Alstyne): Two-sided markets, network externalities, and digital platform regulation.
  • Cryptoeconomics (Buterin, Roughgarden, Huberman): Game theory applied to blockchain protocol design.

Current Landscape (2026)

• The 2024–2026 period is characterised by a widening divergence between capability advances in AI and macroeconomic measurement of AI’s contribution. Frontier models (GPT-4o, Claude Opus 4, Gemini 2.0) perform at or above human expert level on many narrow cognitive benchmarks, yet aggregate TFP statistics in major economies show at most modest improvement. This is consistent with Brynjolfsson’s J-curve but also with Acemoglu’s more pessimistic view that AI’s task coverage remains insufficient for macroeconomic significance.
• Key developments shaping the economic landscape in 2026:
  • Enterprise AI deployment scaling: Major technology firms (Microsoft, Google, Salesforce) report quantifiable productivity gains from AI copilot products in specific workflows — software development (GitHub Copilot), customer service, document processing — but aggregate numbers remain elusive.
  • Labour market resilience: Despite rapid AI capability growth, major economies have not experienced the mass unemployment feared in 2022–2023 AI impact reports. Unemployment rates in the US and UK remain near historic lows in 2024–2025, though sectoral shifts (growth in AI roles, decline in some routine white-collar categories) are accelerating.
  • Agentic AI economics: Autonomous AI agents capable of multi-step task completion (Devin, Claude Computer Use, GPT-4o with tool use) introduce new principal-agent risks and raise novel questions about liability, accountability, and incentive alignment for AI systems acting economically on behalf of users.
  • Digital Market Act (DMA) and DMCC enforcement: EU and UK competition regulators are applying Tirole-style platform economics analysis to impose interoperability, data-portability, and self-preferencing prohibitions on AI-enabled gatekeepers.
  • DAO economic maturity: Major DeFi protocols (Uniswap, Aave, Compound) have accumulated multi-billion-dollar treasuries governed by DAO token holders, providing the first large-scale empirical tests of on-chain mechanism design at macroeconomic scale.
  • Quadratic mechanisms in practice: Gitcoin Grants rounds using CO-QF continue to fund public goods; Optimism’s RetroPGF mechanism distributes protocol revenue retroactively to contributors of public goods, testing alternative mechanisms for open-source sustainability.

UK Context (Imperial / Edinburgh / UCL / Cambridge / Manchester academic; Northern English industrial)

• The United Kingdom has distinctive academic and policy contributions to the economics of AI and digital systems:
  • LSE Centre for Economic Performance (CEP): Europe’s leading policy-oriented economics research centre. CEP’s AI productivity research includes the first representative international survey of firm-level AI use across UK, US, Germany, and Australia (~6,000 CFOs/CEOs), and a May 2024 survey of 373 UK firms with CBI on AI adoption patterns. Nobel Laureate Christopher Pissarides and productivity expert Mary O’Mahony have argued AI can improve job quality and productivity if firms invest in training and wellbeing alongside AI deployment. CEP’s Programme on Innovation and Diffusion (POID) tracks technology diffusion across the UK productivity distribution.
  • Imperial College Business School: Jonathan Haskel (Economics Professor, former MPC member) is a leading theorist of intangible capital — investment in software, R&D, brand, and organisational capital that is systematically under-measured in national accounts but increasingly dominates AI-era firm value. His work with Stian Westlake (“Capitalism Without Capital”, 2017) argues that intangible-intensive economies face distinctive challenges around inequality, secular stagnation, and measurement that require new policy frameworks.
  • Manchester School (economic geography): The University of Manchester’s economic geography tradition (rooted in Alfred Marshall’s industrial district concept) has been applied to AI cluster dynamics and regional productivity divergence. Manchester’s Alliance Manchester Business School hosts research on AI adoption patterns among Northern English SMEs, documenting how AI diffusion reinforces the North-South UK productivity gap unless active industrial policy intervenes.
  • Cambridge Judge Business School: Research on platform economics, digital market competition, and the economics of data — directly informing UK CMA and FCA policy on AI and digital markets.
  • UCL Economics and the Institute for Innovation and Public Purpose (IIPP): Mariana Mazzucato’s IIPP research frames AI as a technology requiring public sector investment and mission-oriented governance to deliver broad social value, challenging the assumption that markets alone will direct AI toward public welfare objectives.
  • Nottingham Centre for Decision Research: Behavioural economics research on AI adoption decision-making, cognitive bias in algorithmic automation choices, and the psychology of principal-agent delegation to AI systems.
  • Northern industrial context: The economic geography of Northern England — former manufacturing heartlands in Sheffield steel, Leeds textiles, Newcastle engineering — makes AI-driven labour market economics acutely politically salient. ONS data shows that occupations with the highest AI exposure are concentrated in London and the South East, while communities dependent on routine manufacturing and services face the displacement effect without the task-creation effect observed in high-skill clusters. UK Shared Prosperity Fund and Levelling Up initiatives interact with AI labour market dynamics in ways still under active economic study.
  • Sheffield and Leeds digital economy: The Northern Powerhouse partnership and Leeds City Region growth agenda have prioritised digital and AI economy development as part of economic rebalancing. The Leeds/Bradford conurbation has attracted fintech and AI firms partly due to lower commercial property costs and a substantial STEM graduate pipeline from University of Leeds, Leeds Beckett, Bradford, Huddersfield, and Sheffield Hallam. The Sheffield Digital cluster and Leeds Digital Festival (Europe’s largest free digital festival) signal emerging AI economy strength, though the productivity gap with London-based tech clusters persists in ONS regional accounts data.
  • Bank of England and monetary economics: The Bank of England’s Centre for Central Banking Studies, its Financial Policy Committee, and Prudential Regulation Authority are actively engaged with AI risk in financial services — AI model risk in credit decisions, large language models in audit and compliance, and AI-driven market instability risks. The BoE’s 2024 Financial Stability Report identified AI-generated correlated behaviour across financial firms as a systemic risk concern. These macroprudential economics concerns link AML KYC Compliance, AI Liability, and CBDC Frameworks in the UK regulatory context.

Future Directions (2026–2030)

• The economics of AI and digital systems faces several major open questions and anticipated developments over 2026–2030:
  • Resolving the productivity paradox: The key empirical question is whether AI productivity gains will appear in aggregate statistics by 2027–2030 as Brynjolfsson’s J-curve predicts. If they do not materialise, more pessimistic frameworks (Acemoglu’s task-coverage argument, Gordon’s secular stagnation thesis) will gain empirical support.
  • Agentic AI labour economics: As AI agents take on longer-horizon tasks with reduced human oversight, the labour-market substitution question shifts from “task automation” to “job automation” — a qualitative change in exposure that existing task-based models may not fully capture.
  • Endogenous task creation: Acemoglu’s framework predicts net positive labour-market outcomes only if AI enables creation of new tasks for human workers. Whether generative AI catalyses new occupational categories (AI trainers, prompt engineers, AI safety auditors, AI economists) in sufficient quantities remains the central empirical test of the technology’s distributional impact.
  • Cryptoeconomic maturation: On-chain mechanism design will face its largest empirical tests as DeFi protocols with billion-dollar treasuries face bear markets, regulatory pressure, and governance crises. The economics of DAO resilience, optimal tokenomics, and sustainable public-goods funding remain unsolved.
  • Sovereign AI and geopolitical economics: Nations pursuing AI sovereignty (EU, UK, China, India) will generate new economic trade-offs between efficiency gains from AI integration and national security costs of technology dependence. The economics of compute control (export controls, strategic reserve compute, national AI clouds) is an emerging sub-field.
  • AI and inequality: The distributional economics of AI — across income groups, occupations, regions, and countries — is the most politically consequential open question. Whether AI is a convergence technology (enabling developing economies to leapfrog) or a divergence technology (widening rich-poor and skilled-unskilled gaps) will determine its legacy. IMF’s “Mind the Gap” framing (2025) suggests divergence is the current base case absent active policy.
  • Quadratic and plural mechanisms at scale: The extension of Weyl’s plurality mechanisms from blockchain public goods to broader democratic governance and resource allocation is an active frontier, with implications for AI governance, urban planning, and global commons management.
  • AI in monetary economics: As AI agents become capable of autonomous financial transactions — purchasing, contracting, paying for API services — questions arise about whether AI agents should hold assets, what monetary instruments are suitable for agent-to-agent payments (stablecoins, CBDC payment rails, Bitcoin Lightning micropayments via Cashu), and how existing monetary policy transmission channels function when a significant fraction of transactions involve non-human agents.
  • New labour economics frameworks: The task-based model (Acemoglu-Restrepo) and the worker-centred model (Autor) may both require extension as AI moves from task automation to holistic job performance — where AI systems handle entire job roles rather than discrete tasks. This changes the unit of analysis from task exposure to job exposure, requiring new empirical frameworks and retraining policy designs.
  • Open-source AI economics: The economic sustainability of open-source foundation models (Meta LLaMA, Mistral, Falcon) depends on mechanisms for funding continued maintenance, safety work, and capability improvements without the revenue stream of proprietary API providers. This is a public goods problem in Weyl’s sense: open models are non-excludable (anyone can download weights) and non-rival, creating classical free-rider dynamics. Novel mechanisms being explored include: open collective funding, foundation model cooperatives, retroactive public goods funding, and compute donations from hyperscalers with strategic interests in a healthy open ecosystem.
  • Automated mechanism design: As AI systems themselves become capable of designing economic mechanisms — generating auction formats, contract structures, or token distribution schemes — the field of mechanism design is being transformed. AI-assisted mechanism design may discover novel incentive-compatible rules beyond what human economists can derive analytically, but raises verification challenges: how do we audit AI-designed mechanisms for hidden failure modes, manipulability, or distributional unfairness?

Information Economics and Digital Markets

Non-Rival Goods and Zero-Marginal-Cost Economics

• A defining characteristic of digital products — software, trained AI models, data, digital media — is that they are non-rival: consumption by one party does not diminish availability for others. This contrasts with traditional economic goods (coal, land, labour hours) where scarcity drives price above zero. When the marginal cost of reproduction approaches zero (as with digital downloads, API calls served by already-trained models, or on-chain data retrieval), standard competitive equilibrium theory predicts prices will converge to zero in the long run. The implication for AI economics is profound: if foundation models are non-rival and can be replicated at near-zero marginal cost, competitive market pressures should drive prices toward zero unless providers can sustain differentiated positions through network effects, proprietary data, integration lock-in, or performance leadership.
• This dynamic is already visible in the AI model pricing collapse of 2023–2025: GPT-4’s launch price of $0.06/1Kinputtokensdeclinedtobelow$0.001 for comparable-quality open-weight models by mid-2025, following a price decay trajectory faster than Moore’s Law. The economics of non-rival AI models thus resemble the economics of software more than the economics of hardware or services: fixed costs of model training are substantial ($10M–$100B+ for frontier models), but marginal costs of inference are converging toward commodity compute pricing. This creates first-mover advantage dynamics where early capability leaders can earn quasi-rents during the period before competitive catch-up, but face ongoing commoditisation pressure.
• Paul Romer’s endogenous growth theory explicitly incorporates non-rivalry: in his 1990 model, ideas (non-rival) are the engine of growth, producing increasing returns at the aggregate level even under diminishing returns at the individual firm level. AI models — as codified, reproducible algorithms — are the most economically significant non-rival good in history, potentially accelerating Romer’s knowledge-driven growth dynamic.

Network Effects and Platform Tipping

• Digital markets frequently exhibit network externalities — the value of a product to a user increases with the number of other users (Katz and Shapiro 1985). Network effects create winner-take-most dynamics in platform markets: once a platform achieves critical mass, users face coordination problems migrating to alternatives, creating lock-in that insulates incumbents from competition. AI markets exhibit several distinct network-effect types:
  • Direct network effects (user-to-user): Social platforms (LinkedIn, X/Twitter) where AI assistants are embedded acquire value from the underlying social graph that pure AI startups cannot replicate.
  • Data network effects (indirect): More users generate more data, enabling better model training, producing better models that attract more users. The strength of this effect is contested empirically — Lambrecht and Tucker (2019) find data advantages in advertising erode quickly as datasets scale beyond a threshold — but for specialised domains (medical AI, financial AI) with scarce labelled data, data network effects may be more durable.
  • Ecosystem network effects: AI platforms (OpenAI API, Google Vertex AI) benefit from growing ecosystems of third-party developers, fine-tuned models, and integrations that increase switching costs for enterprise customers. This is the most strategically significant network effect for AI competition analysis.
• Platform tipping — when one platform achieves dominant position that becomes self-reinforcing — has been documented in search (Google 90%+ global share), mobile OS (iOS/Android duopoly), and social media (Meta’s successive tipping across Facebook, Instagram, WhatsApp). The economic question for AI is whether foundation model APIs, AI coding tools, or AI enterprise platforms will exhibit similar tipping dynamics that justify early regulatory intervention under competition law.

Adverse Selection, Signalling, and AI Labour Markets

• Akerlof’s market for lemons (1970) demonstrated that information asymmetry about product quality can cause markets to unravel: sellers know more about quality than buyers, buyers rationally discount prices to account for average quality, sellers of high-quality goods exit, quality falls further, prices fall further — a death spiral. Michael Spence’s signalling theory (1973) provided the complementary insight that in labour markets, education functions primarily as a signal of pre-existing ability rather than a productivity investment — a signalling equilibrium sustained because the cost of obtaining the signal is lower for high-ability workers.
• AI reshapes these dynamics in several ways:
  • Credential devaluation: If AI can perform the tasks for which educational credentials were a proxy signal (legal research, basic coding, document drafting), the equilibrium returns to education-as-signal may decline. Spence’s signalling model predicts this shifts equilibria toward alternative signals (demonstrated AI capability, portfolio-based assessment, AI-verified skill certifications).
  • AI hiring tools as screening mechanisms: AI-powered recruitment screening reduces employers’ search costs but may introduce algorithmic bias as a new form of adverse selection — systematically screening out candidates from protected groups in ways that are statistically discriminatory even when individually rational. The Algorithmic Bias and Variance dynamics interact directly with labour economics here.
  • Gig economy and AI work platforms: Platforms like Scale AI, Turing, and Mechanical Turk mediate labour markets for AI data annotation and evaluation tasks. Information asymmetries about worker quality, task completion, and output accuracy in these markets replicate classic Akerlof dynamics at scale.

Behavioural Economics of AI Adoption

Cognitive Biases in Technology Adoption Decisions

• Standard economic models assume rational optimising agents who correctly estimate the returns to technology adoption and invest up to the point where marginal returns equal marginal costs. Behavioural economics (Kahneman and Tversky’s Prospect Theory, Thaler’s nudge framework) documents systematic deviations from this ideal: present bias leads firms to underinvest in technologies with delayed payoffs (consistent with J-curve dynamics), loss aversion makes managers reluctant to restructure workflows around AI when disruption costs are salient, and status quo bias creates organisational inertia against AI adoption even when the expected value calculation favours it.
• These behavioural failures are particularly acute for AI adoption because:
  • Benefits are diffuse and deferred (productivity gains visible in 2–5 years, after organisational restructuring) while costs are concentrated and immediate (implementation costs, retraining disruption, uncertainty about outcomes).
  • AI outcomes are ambiguous and hard to attribute — it is difficult to tell whether a productivity improvement came from AI or from accompanying process changes.
  • Automation bias (over-reliance on algorithmic outputs) and algorithm aversion (under-reliance after seeing a single algorithmic failure) are documented biases that create sub-optimal human-AI teaming equilibria.
• The Nottingham Centre for Decision Research has produced empirical work on AI delegation decisions — when workers choose to use AI assistance, override AI recommendations, or avoid AI systems entirely — finding that framing effects and loss aversion significantly influence adoption behaviour independent of objective AI capability.

Fairness Preferences and Redistributive Economics

• Experimental economics (Fehr and Gächter ultimatum and public goods games) documents that humans have strong fairness preferences that deviate from pure self-interest. These preferences shape political economy of AI: even if AI raises aggregate GDP, distributional consequences (widening wage inequality, regional displacement of industrial communities) can trigger political resistance that limits policy space for AI investment. The economics of AI governance must therefore account for distributional constraints — policies that are Pareto-efficient in aggregate but produce concentrated losers will face political opposition regardless of net welfare calculus.
• In cryptoeconomics, fairness preferences manifest as resistance to token allocations perceived as extractive — large founder/investor pre-mines, retroactive airdrops that exclude early users, or governance changes that dilute existing token holders. The legitimacy of a protocol’s economic design affects adoption and retention in ways that purely efficiency-maximising token models fail to capture.

Economic Analysis of AI Safety and Alignment

Externalities and Market Failures in AI Development

• AI development exhibits several negative externalities — costs imposed on third parties not reflected in the private cost-benefit calculation of AI developers:
  • Labour market displacement externalities: Firms that automate gain private productivity benefits, but the social cost of worker displacement (unemployment, retraining costs, community disruption) is borne by workers, governments, and communities rather than by the automating firm. This produces over-automation relative to the social optimum — a textbook Pigouvian externality case.
  • Safety externalities: An AI developer who deploys an unsafe system bears private liability costs only if those can be attributed and enforced; the broader risks (systemic disruption, erosion of trust in digital systems, catastrophic-risk tail scenarios) are public bads. This produces under-investment in safety relative to the social optimum.
  • Data privacy externalities: AI training on personal data without adequate consent imposes privacy costs on data subjects. The EU GDPR and UK DPAA attempt to internalise these costs through compliance requirements, but enforcement gaps mean significant external costs remain.
  • Environmental externalities: AI training compute has substantial carbon footprint. Carbon Footprint Measurement and Carbon Credit Tracking pages address the market-design tools (carbon pricing, cap-and-trade) for internalising these externalities.
• The Pigouvian taxation framework and Coasean bargaining framework offer competing approaches to externality correction. Where transaction costs are low (small numbers of affected parties, clear property rights), Coasean bargaining may achieve efficient outcomes without government intervention. Where transaction costs are high (millions of affected workers, diffuse harm), Pigouvian taxes, liability rules, or regulatory mandates are required. AI externalities fall predominantly in the latter category, justifying regulatory intervention through frameworks like the EU AI Act and UK AI Governance White Paper.

AI Alignment as a Public Goods Problem

• AI safety and alignment research exhibits the economic structure of a public good: it is non-excludable (safety knowledge, once produced, benefits all AI developers) and non-rival (one organisation’s use of safety techniques does not reduce availability for others). Classical public goods theory predicts under-provision by private markets, because free-rider incentives lead each developer to under-invest in safety knowledge production, relying on others to bear the cost. This is the economic justification for public funding of AI safety research (UK AISI, US NIST AI Safety Institute, Anthropic’s Constitutional AI research), prizes for safety advances, and open publication norms in safety-relevant areas.
• Quadratic Funding (Weyl, Buterin, Hitzig 2019) was explicitly designed for this use case: funding open-source software and safety-relevant public goods that private markets under-supply. Gitcoin’s deployment of QF for Ethereum ecosystem public goods is a live experiment in whether mechanism design can solve the AI/crypto public goods problem at scale.

Welfare Economics and Distributional Effects of AI

Measuring AI Welfare: Consumer Surplus and Digital Surplus

• Standard welfare economics measures social benefit as the sum of consumer surplus (value consumers receive above the price paid) and producer surplus (profit above cost). For digital and AI goods where marginal cost approaches zero and large portions are provided free (Google Search, free AI tiers), traditional surplus measures fail dramatically. The price paid for a service like Google Maps is zero, yet the consumer surplus from not needing to purchase a physical map, navigation device, and geographic knowledge may be $1,000+ per year per user. GDP statistics that count only market transactions systematically miss this surplus.
• Brynjolfsson has developed willingness-to-accept (WTA) surveys to measure digital surplus directly — asking users what compensation they would require to give up access to specific digital services for one month. Estimates reveal massive consumer surplus: Facebook users require ~$40/month;GoogleMaps$3,600/year; email ~$8,400/year.ThetotaldigitalsurplusunmeasuredinUSGDPwasestimatedatapproximately\ast \ast$1.1 trillion/year** in 2019 (Brynjolfsson et al., GDP-B measure). For AI specifically, similar surveys will need to be developed as AI capabilities become embedded in daily workflows — the measured GDP impact of AI will systematically understate welfare impact.
• Intangible capital measurement (Corrado, Hulten, and Sichel 2009; Haskel and Westlake 2017) provides a complementary approach: treating firm investments in AI capabilities, software, data assets, and organisational restructuring as capital formation rather than expenses. The UK ONS and BEA have begun incorporating expanded intangible capital categories into national accounts, partially correcting the measurement gap.

Distributional Welfare: Gini Coefficient Dynamics Under AI

• The Gini coefficient and related inequality measures are sensitive to the distributional concentration of AI benefits. If AI primarily augments high-skill workers (who already command high wages), reduces costs of AI-produced goods most enjoyed by high-income households, and concentrates ownership of AI-producing capital in already-wealthy hands, AI will exacerbate wealth and income inequality. Current evidence leans in this direction:
  • Capital-labour income split: Acemoglu’s framework predicts AI widens the gap between capital income (returns to AI-producing firms and their shareholders) and labour income (wages). US corporate profit share of GDP has risen from ~9% in 2000 to ~12% in 2024, partly attributable to software and AI automation.
  • Superstar firm dynamics: Autor, Dorn, Katz, Patterson, and Van Reenen (2020) document rising concentration of economic activity in “superstar firms” — hyper-productive companies that capture disproportionate market share. AI is expected to amplify this dynamic: AI-enabled productivity advantages compound over time, and AI capacity is concentrated in the largest firms with greatest training data and compute resources.
  • Geographic inequality: AI production (Silicon Valley, London, Beijing) and AI-enabled high-wage employment are geographically concentrated, exacerbating regional inequality. This is particularly acute in the UK, where the North-South productivity divide predates AI but may be amplified by differential AI adoption rates.
• Counter-arguments exist: AI may reduce inequality by automating high-cost services (legal advice, medical triage, tutoring) that are currently affordable only to high-income households, democratising access to quality services. AI Adoption in healthcare, education, and legal services could produce significant progressive distributional effects if access is broad and affordable. The net effect remains empirically contested.

Economics of Multi-Agent Systems

Market Coordination in Multi-Agent AI Ecosystems

• As CLI Multi-Agent Systems and Agent Frameworks mature, economic coordination between multiple autonomous AI agents emerges as a distinct domain. Classical economics studies how markets coordinate the actions of large numbers of self-interested human agents through price signals; the analogous question for AI multi-agent systems is how coordination protocols (message-passing, shared memory, market mechanisms, voting) can align multiple AI agents toward collective objectives without central control. Key economic sub-questions include:
  • Resource allocation among agents: When multiple agents share compute, memory, or API rate limits, how should these scarce resources be allocated? Auction mechanisms (first-price, second-price, VCG) provide theoretically well-studied approaches; token-based resource accounting (giving each agent a budget of compute tokens to spend) creates implicit markets.
  • Agent specialisation and trade: In swarm architectures where different agents specialise in sub-tasks (research, coding, verification, writing), the gains from specialisation — Adam Smith’s division of labour — apply. Economic analysis of optimal specialisation depth, communication costs, and coordination mechanisms for AI swarms draws directly on classical comparative advantage theory.
  • Emergent market behaviours: When AI agents interact repeatedly and can adapt strategies, emergent market-like behaviours (price discovery, collusion, predatory behaviour) may arise. The economics of agent interaction at scale requires both mechanism design (designing rules that prevent collusion and predation) and empirical observation of emergent equilibria.
  • Principal hierarchies: Enterprise AI deployments involve nested principal-agent structures — employee (human) → manager (human) → enterprise (human organisation) → AI agent → AI sub-agents — each level introducing agency costs. The economics of vertical principal hierarchies with AI agents requires extending classical Williamson transaction-cost hierarchies to include software agents as organisational nodes.

Comparative Economics: AI vs. Previous General-Purpose Technologies

Historical GPT Productivity Analysis

• The economic history of General-Purpose Technologies (GPTs) — technologies with broad applicability, scope for improvement, and innovation complementarities across the economy — provides the empirical baseline against which AI’s economic impact should be assessed. Economists Bresnahan and Trajtenberg (1995) formalised the GPT concept; subsequent empirical work (Jovanovic and Rousseau 2005) estimated the economic contribution of electricity, the steam engine, ICT, and the internet.
• Key historical data points:
  • Electricity: Adopted from the 1880s; productivity gains in US manufacturing peaked in the 1920–1929 period, approximately 40 years after widespread adoption, as firms restructured factories from shaft-drive to decentralised electric motors. Annual TFP growth in electrified manufacturing reached 5–6% during this period.
  • Information and Communications Technology (ICT): Widely adopted from the 1970s; productivity paradox documented in the 1980s (Solow 1987); acceleration in US TFP growth visible from 1995–2005, approximately 20–30 years after widespread adoption. The productivity acceleration of 1995–2000 (2.5–3% annual TFP growth vs. 1.4% prior decade) was the most significant technology-driven productivity surge since electrification.
  • Internet: Structural deployment from 1993; business model maturation 1998–2008; productivity impacts still being absorbed. E-commerce reached 20%+ of US retail only in 2020, 25 years after the web’s commercial launch.
• Brynjolfsson’s J-curve prediction is that AI’s productivity surge is 10–15 years away from the 2020 GPT-3 breakthrough moment, implying the boom should be visible in 2030–2035 statistics. Acemoglu’s more sceptical view holds that current AI lacks the breadth of task automation needed to drive GPT-scale productivity growth absent fundamental changes in AI capability and business adoption patterns.

Unique Features of AI as a GPT

• AI differs from prior GPTs in several economically significant ways:
  • Self-referential improvement: AI can be applied to AI research itself — automating hypothesis generation, literature review, code implementation, and experiment design. This creates potential recursive acceleration absent in electricity or ICT. However, fundamental physical and epistemic limits (alignment failures, distribution shift, hallucination) impose constraints on this recursion.
  • Cognitive rather than physical automation: Previous automation GPTs (steam, electricity, robotics) primarily displaced physical labour; AI primarily displaces cognitive labour. This is economically significant because cognitive labour commands higher wages, making AI potentially more economically disruptive per displaced task-hour, but also because cognitive labour is more heterogeneous and context-dependent, making full automation harder.
  • Speed of improvement: AI capability has improved at unprecedented speed (GPT-2 to GPT-4 in four years; a roughly 100× improvement in benchmark performance). This compresses the historical J-curve timeline and may reduce the complementary-investment lag if organisational adaptation can occur faster.
  • Global simultaneity: AI is deployed globally essentially simultaneously, unlike electricity (which diffused across decades and countries). This compresses geographic productivity divergence timelines but also means distributional consequences emerge faster.

Cryptoeconomics: Stablecoins, CBDCs, and Monetary Economics

The Monetary Economics of Stablecoins

• Stablecoins — digital assets designed to maintain a fixed peg against a reference currency or asset — represent the most direct intersection of AI/blockchain economics and monetary economics. Their economics diverge sharply by design:
  • Fiat-backed stablecoins (USDC, USDT): Each token is backed by reserves of fiat currency or short-term government securities held in custodial accounts. The economics resemble narrow banking: the issuer earns interest on reserves while token holders receive zero yield (pre-2023) or yield-sharing (post-regulatory pressure). Reserve transparency, audit quality, and custodian solvency are the key risk factors. USDT’s ~$100B market cap makes Tether effectively a significant participant in short-term US Treasury markets.
  • Algorithmic stablecoins (TerraUSD/LUNA — collapsed May 2022): Attempted to maintain peg through arbitrage incentives rather than collateral, relying on seigniorage from a companion volatile token (LUNA) to absorb demand shocks. The economic design was fatally flawed — under conditions of simultaneous sell pressure on both UST and LUNA, the arbitrage mechanism was unable to maintain the peg, producing a death spiral that destroyed ~$60B in market value in 72 hours. This is a landmark case study in failed mechanism design: the stabilisation mechanism assumed positive-sum conditions that do not hold in a bank-run equilibrium.
  • Crypto-collateralised stablecoins (DAI): Over-collateralised by crypto assets (ETH, WBTC) with automated liquidation triggers. The economics require maintaining collateralisation ratios above 150%; protocol-governed stability fees and DSR (DAI Savings Rate) function as monetary policy instruments. MakerDAO’s governance token (MKR) is the residual claimant on seigniorage, creating principal-agent tensions between MKR holders and DAI users.
• The CBDCs page extends this analysis to central-bank-issued digital currency, where the monetary economics involves disintermediation risk to commercial banks, programmable money policy, and cross-border settlement efficiency.

Financial Economics: DeFi and Automated Market Makers

• Decentralised Finance (DeFi) replaces traditional financial intermediaries with smart-contract-based protocols. The economics of key DeFi primitives:
  • Automated Market Makers (AMMs): Uniswap’s constant-product formula (x × y = k) creates a continuous on-chain exchange with no order book. The economics involve: liquidity provider (LP) returns from trading fees minus impermanent loss (the opportunity cost of holding pool assets vs. holding them directly, which arises when asset prices diverge). LP economics are often negative on a risk-adjusted basis for volatile pools, explaining why DeFi liquidity relies heavily on token incentive subsidies — raising sustainability questions once subsidies decline.
  • Lending protocols (Aave, Compound): Over-collateralised lending with algorithmically determined interest rates responding to supply-demand for each asset. The economics resemble money markets with automated collateral calls (liquidations) replacing credit officers. Liquidation mechanics — who can liquidate, at what discount, with what gas-cost economics — are critical to protocol solvency and are themselves a mechanism design problem.
  • Yield farming and liquidity mining: Protocols distribute governance tokens to liquidity providers as subsidy, temporarily raising LP returns above opportunity cost to bootstrap liquidity network effects. The economics are explicitly Ponzi-adjacent during early phases: returns are funded by token inflation rather than economic value creation. Sustainable protocols must transition from token-subsidised to fee-sustained liquidity before token inflation dilutes governance value to zero.

Economics of AI Governance and Regulation

Regulatory Economics: Capture Theory and AI

• George Stigler’s regulatory capture theory (1971) argues that regulated industries tend to capture their regulators over time, using regulatory machinery to erect barriers to entry and protect incumbent profits rather than serve the public interest. Applied to AI regulation, this concern motivates: (a) fears that incumbent AI hyperscalers (Google, Microsoft, OpenAI) will shape AI regulation to impose compliance costs that deter competitive entry; (b) concerns that safety-focused AI regulation may inadvertently function as protectionist industrial policy; (c) the design of regulatory institutions (UK AISI, EU AI Office) to resist capture through independence, rotation, and transparency requirements.
• The alternative “public interest” theory of regulation (Pigou, Arrow, Samuelson) holds that regulation addresses genuine market failures (externalities, information asymmetry, market power). Both theories find support in AI regulatory practice: the EU AI Act’s high-risk classification system has genuine safety rationale, but its compliance architecture clearly favours large incumbents over open-source and start-up entrants.

The Economics of AI Liability

• The AI Liability framework interacts with economics through several channels. Tort liability for AI-caused harm (medical misdiagnosis, discriminatory lending decisions, autonomous vehicle accidents) functions as a Pigouvian instrument: by forcing AI deployers to internalise harm costs, liability creates incentives for safety investment. However, liability rules must be designed carefully to avoid:
  • Over-deterrence: If liability costs are high and uncertain, deployers may under-deploy beneficial AI (conservative bias) — a welfare loss if AI benefits exceed harms.
  • Judgment-proof problem: Small AI developers may lack assets sufficient to pay damages, eliminating deterrence for under-capitalised firms while leaving the incentive in place for large well-capitalised ones.
  • Attribution difficulty: When AI systems produce harm through multi-step pipelines (data → model → API → application → user), attributing liability across the chain requires novel legal economics frameworks not yet established in UK or EU law.
• The UK Law Commission and EU Product Liability Directive (revised 2024) are actively developing liability frameworks for AI. Economic analysis of these frameworks — impact on innovation incentives, insurance market development, distributional effects of different liability rules — is a rapidly growing applied economics subfield.

Competition Economics and the Digital Markets Act

• The EU’s Digital Markets Act (DMA) (effective March 2024) and the UK’s Digital Markets, Competition and Consumers Act (DMCC) (effective January 2025) represent the most significant application of competition economics to digital and AI markets in a generation. Both frameworks draw on Tirole-Rochet two-sided market theory to identify gatekeepers (platforms controlling access to large user bases that must be accessible to business users) and impose ex ante regulation — structural rules imposed before harm is demonstrated, rather than traditional ex post antitrust enforcement after market power is established. The economic rationale: in digital markets characterised by tipping dynamics and high switching costs, by the time ex post remedies are applied, competitive harm is already entrenched. Designated gatekeepers under DMA include Google, Apple, Meta, Amazon, and Microsoft; the UK CMA’s DMCC strategic market status designations are expected in 2025–2026. The economic question under active research is whether these regulatory frameworks impose welfare-improving constraints or risk stifling legitimate investment and innovation — a tension that empirical economists are beginning to evaluate as enforcement decisions accumulate.

AI and Financial Markets Economics

Asset Pricing and AI

• Financial economics examines how AI affects asset prices, market microstructure, and investor behaviour. AI-driven algorithmic trading now accounts for a majority of equity market volume in developed markets, with implications for price discovery, market liquidity, and systemic risk. High-frequency trading (HFT) algorithms exploit microsecond informational advantages; large language model-based investment research distils earnings calls and news into trading signals; reinforcement learning algorithms optimise portfolio allocation across thousands of assets. The economic consequences include:
  • Market efficiency gains: AI aggregates dispersed information faster, moving prices toward fundamental values more rapidly. The efficient market hypothesis (Fama 1970) is better approximated when AI processes public information with low latency.
  • New forms of market instability: Flash crashes (May 2010, August 2015) demonstrated that coordinated AI trading responses to identical signals can create correlated selling cascades that temporarily disconnect prices from fundamentals. AI-induced correlations across asset classes create new systemic risk patterns not well-captured by existing regulatory models.
  • Alpha decay: As AI trading strategies proliferate, the informational edges they exploit are arbitraged away faster. This compresses the duration of alpha — excess returns — from months to weeks or days, accelerating the competitive treadmill in asset management.
• The economics of AI-generated financial advice raises AI Liability and AML KYC Compliance issues: when AI robo-advisors recommend investment portfolios, the principal-agent relationship (client, robo-advisor, model provider, broker-dealer) creates complex fiduciary duty and liability attribution questions.

Prediction Markets and Information Aggregation

• Prediction markets (Hanson’s Combinatorial Information Markets, Augur, Polymarket) aggregate dispersed private information into probability estimates through price signals — a mechanism for information elicitation that Robin Hanson has proposed as a general governance tool (Futarchy). Blockchain-based prediction markets improve on centralised equivalents by removing the trusted intermediary, but introduce smart-contract execution risk. The economics of prediction markets connect to AI in two ways: AI systems can serve as traders that aggregate information efficiently (or exploit prediction market prices to make decisions), and prediction markets can be used to evaluate AI system claims (e.g., forecasting AI capability timelines or safety incident probabilities). Polymarket’s growth to $1B+ monthly volume by 2025 has made on-chain prediction markets a meaningful economic institution for information aggregation.

Economics of Data and Intellectual Property

Data as an Economic Asset

• Data occupies a strange position in economic theory: it is non-rival (one party’s use does not preclude others’) but is conventionally treated as exclusive (property rights are asserted through contractual restriction, technical access controls, and IP law). This creates tension: economic efficiency favours broad data sharing (non-rivalry means no opportunity cost to sharing), while private incentives to produce data require the ability to monetise it (requiring exclusivity). AI has intensified this tension dramatically: training large foundation models requires vast corpora of text, images, and code; much of this data was produced without anticipation of being used as AI training input; and the economic value of the trained models is uncertain at the time of data collection.
• The markets for data (Agrawal, Gans, and Goldfarb 2018 “Prediction Machines”) framework treats AI as a technology that reduces the cost of prediction, and data as the input to prediction machines. Data markets exhibit classic information-economics pathologies: adverse selection (data sellers know more about data quality than buyers), hold-up problems (once data is shared, the value is revealed and re-negotiation is impossible), and network externalities (data value depends on complementary datasets from other parties). These dynamics partly explain why data markets remain thin and illiquid relative to their economic potential, and why AI training has occurred primarily through voluntary data commons (Wikipedia, Common Crawl, GitHub, Reddit) rather than functioning markets.
• UK and EU regulation (GDPR, Data Act, forthcoming UK Data Use and Access Bill) increasingly attempt to create structured data-sharing obligations that reduce transaction costs in data markets while preserving privacy and competitive neutrality.

Research & Literature

Seminal Theoretical Works

• • Coase, Ronald H. “The Nature of the Firm.” Economica 4.16 (1937): 386-405. Transaction cost theory of the firm foundational to DAO economics analysis.
• • Coase, Ronald H. “The Problem of Social Cost.” Journal of Law and Economics 3 (1960): 1-44. Coase Theorem and property rights framework.
• • Arrow, Kenneth J. “Uncertainty and the Welfare Economics of Medical Care.” American Economic Review 53.5 (1963): 941-973. Information asymmetry and market failure.
• • Mirrlees, James A. “An Exploration in the Theory of Optimum Income Taxation.” Review of Economic Studies 38.2 (1971): 175-208. Foundational principal-agent / optimal contracting theory.
• • Holmström, Bengt. “Moral Hazard and Observability.” Bell Journal of Economics 10.1 (1979): 74-91. Canonical principal-agent formulation.
• • Tirole, Jean and Jean-Charles Rochet. “Platform Competition in Two-Sided Markets.” Journal of the European Economic Association 1.4 (2003): 990-1029. Two-sided platform economics.
• • Hart, Oliver and John Moore. “Incomplete Contracts and Renegotiation.” Econometrica 56.4 (1988): 755-785. Incomplete contracts theory.
• • Romer, Paul M. “Endogenous Technological Change.” Journal of Political Economy 98.5 (1990): S71-S102. Non-rival knowledge and endogenous innovation.

Key AI Economics Research

• • Brynjolfsson, Erik and Andrew McAfee. The Second Machine Age: Work, Progress, and Prosperity in a Time of Brilliant Technologies. W.W. Norton, 2014. General-purpose technology framework.
• • Brynjolfsson, Erik, Daniel Rock, and Chad Syverson. “Artificial Intelligence and the Modern Productivity Paradox: A Clash of Expectations and Statistics.” NBER Working Paper No. 24001, 2017. https://www.nber.org/papers/w24001
• • Brynjolfsson, Erik, Daniel Rock, and Chad Syverson. “The Productivity J-Curve: How Intangibles Complement General Purpose Technologies.” American Economic Journal: Macroeconomics 13.1 (2021): 333-372.
• • Brynjolfsson, Erik, Avinash Collis, W. Erwin Diewert, Felix Eggers, and Kevin J. Fox. “GDP-B: Accounting for the Value of New and Free Goods in the Digital Economy.” NBER Working Paper No. 25695 (2019). Measuring digital consumer surplus.
• • Acemoglu, Daron and Pascual Restrepo. “Automation and New Tasks: How Technology Displaces and Reinstates Labor.” Journal of Economic Perspectives 33.2 (2019): 3-30.
• • Acemoglu, Daron and Pascual Restrepo. “Robots and Jobs: Evidence from US Labor Markets.” Journal of Political Economy 128.6 (2020): 2188-2244.
• • Acemoglu, Daron, Simon Johnson, and Pascual Restrepo. Power and Progress: Our Thousand-Year Struggle Over Technology and Prosperity. PublicAffairs, 2023.
• • Acemoglu, Daron. “The Simple Macroeconomics of AI.” NBER Working Paper No. 32487, April 2024. Predicts <0.66% TFP gain over 10 years. https://www.nber.org/papers/w32487
• • Autor, David, Levy, Frank and Murnane, Richard J. “The Skill Content of Recent Technological Change: An Empirical Exploration.” Quarterly Journal of Economics 118.4 (2003): 1279-1333. Task framework for labour economics.
• • Autor, David, David Dorn, Lawrence F. Katz, Christina Patterson, and John Van Reenen. “The Fall of the Labor Share and the Rise of Superstar Firms.” Quarterly Journal of Economics 135.2 (2020): 645-709.
• • Aghion, Philippe, Benjamin Jones, and Charles Jones. “Artificial Intelligence and Economic Growth.” NBER Working Paper No. 23928, 2017.
• • Agrawal, Ajay, Joshua Gans, and Avi Goldfarb. Prediction Machines: The Simple Economics of Artificial Intelligence. Harvard Business Review Press, 2018.
• • Fehr, Ernst and Simon Gächter. “Cooperation and Punishment in Public Goods Experiments.” American Economic Review 90.4 (2000): 980-994. Fairness preferences in mechanism design.
• • Katz, Michael L. and Carl Shapiro. “Network Externalities, Competition, and Compatibility.” American Economic Review 75.3 (1985): 424-440. Network effects foundational paper.
• • Rochet, Jean-Charles and Jean Tirole. “Two-Sided Markets: A Progress Report.” RAND Journal of Economics 37.3 (2006): 645-667. Extended two-sided market theory with applications to interchange fees and platform competition.
• • Williamson, Oliver E. “Transaction-Cost Economics: The Governance of Contractual Relations.” Journal of Law and Economics 22.2 (1979): 233-261. Transaction cost theory and vertical integration.
• • North, Douglass C. Institutions, Institutional Change and Economic Performance. Cambridge University Press, 1990. Formal and informal institutions as determinants of economic outcomes.
• • Stiglitz, Joseph E. “The Theory of ‘Screening,’ Education, and the Distribution of Income.” American Economic Review 65.3 (1975): 283-300. Screening and information asymmetry in labour markets.
• • Varian, Hal R. “Markets for Information Goods.” Bank of Japan Conference Paper, 1998. Non-rivalry, versioning, and pricing of digital goods.
• • Hanson, Robin. “Shall We Vote on Values, But Bet on Beliefs?” George Mason University Working Paper, 2007. Futarchy prediction markets as governance mechanism.
• • Farrell, Joseph and Carl Shapiro. “Dynamic Competition with Switching Costs.” RAND Journal of Economics 19.1 (1988): 123-137. Lock-in economics applied to platform competition.
• • Acemoglu, Daron and Pascual Restrepo. “The Wrong Kind of AI? Artificial Intelligence and the Future of Labour Demand.” Cambridge Journal of Regions, Economy and Society 13.1 (2020): 25-35.
• • Olson, Mancur. The Logic of Collective Action: Public Goods and the Theory of Groups. Harvard University Press, 1965. Foundational analysis of free-rider dynamics in collective action, foundational to DAO governance analysis.
• • Bresnahan, Timothy F. and Manuel Trajtenberg. “General Purpose Technologies: ‘Engines of Growth’?” Journal of Econometrics 65.1 (1995): 83-108. GPT framework applied to historical economic transformations.

Mechanism Design and Crypto-Economics

• • Buterin, Vitalik, Zoë Hitzig, and E. Glen Weyl. “A Flexible Design for Funding Public Goods.” Management Science 65.11 (2019): 5171-5187. Quadratic Funding.
• • Weyl, E. Glen. “A Price Theory of Multi-Sided Platforms.” American Economic Review 100.4 (2010): 1642-1672.
• • Roughgarden, Tim. “Transaction Fee Mechanism Design for the Ethereum Blockchain: An Economic Analysis of EIP-1559.” Proceedings of the 22nd ACM Conference on Economics and Computation, 2021.
• • Huberman, Gur, Jacob Leshno, and Ciamac Moallemi. “Monopoly Without a Monopolist: An Economic Analysis of the Bitcoin Payment System.” Review of Economic Studies 88.6 (2021): 3011-3040.
• • Biais, Bruno, Christophe Bisière, Matthieu Bouvard, and Catherine Casamatta. “The Blockchain Folk Theorem.” Review of Financial Studies 32.5 (2019): 1662-1715.
• • Haskel, Jonathan and Stian Westlake. Capitalism Without Capital: The Rise of the Intangible Economy. Princeton University Press, 2017.

International Institution Reports

• • IMF. “The Global Impact of AI: Mind the Gap.” World Paper WP/25/76, April 2025. Gap between AI-frontier and developing economies.
• • Goldman Sachs. “The Potential Long-Run Impact of AI on the US Labor Market.” Research Report, 2023. 15% US GDP uplift projection.
• • OECD. “Artificial Intelligence and the Changing Demand for Skills in the Labour Market.” OECD Publications, April 2024.
• • OECD. “Generative AI and the Future of Work: Global Dialogue.” GPAI Report, January 2025.
• • UK Competition and Markets Authority (CMA). “Foundation Models: Initial Report.” 2024. Platform economics and market concentration concerns.

Provenance

• Acemoglu, Daron. “The Simple Macroeconomics of AI.” NBER Working Paper No. 32487 (2024). https://www.nber.org/papers/w32487
• Brynjolfsson, Erik, Daniel Rock, and Chad Syverson. “Artificial Intelligence and the Modern Productivity Paradox.” NBER Working Paper No. 24001 (2017). https://www.nber.org/papers/w24001
• Brynjolfsson, Erik, Daniel Rock, and Chad Syverson. “The Productivity J-Curve.” AEJ: Macroeconomics 13.1 (2021).
• Acemoglu, Daron and Pascual Restrepo. “Automation and New Tasks.” Journal of Economic Perspectives 33.2 (2019).
• Acemoglu, Daron and Pascual Restrepo. “Robots and Jobs.” Journal of Political Economy 128.6 (2020).
• Acemoglu, Simon Johnson, Pascual Restrepo. Power and Progress (2023).
• Buterin, Vitalik, Zoë Hitzig, E. Glen Weyl. “A Flexible Design for Funding Public Goods.” Management Science 65.11 (2019). https://arxiv.org/abs/1809.06421
• Coase, Ronald H. “The Nature of the Firm.” Economica 4.16 (1937).
• Coase, Ronald H. “The Problem of Social Cost.” Journal of Law and Economics 3 (1960).
• Hart, Oliver and John Moore. “Incomplete Contracts and Renegotiation.” Econometrica 56.4 (1988).
• Tirole, Jean and Jean-Charles Rochet. “Platform Competition in Two-Sided Markets.” JEEA 1.4 (2003).
• Holmström, Bengt. “Moral Hazard and Observability.” Bell Journal of Economics 10.1 (1979).
• IMF. “The Global Impact of AI: Mind the Gap.” WP/25/76 (April 2025). https://www.imf.org/-/media/files/publications/wp/2025/english/wpiea2025076-print-pdf.pdf
• Goldman Sachs. “The Potential Long-Run Impact of AI on the US Labor Market” (2023).
• OECD. “AI and the Changing Demand for Skills in the Labour Market” (April 2024). https://www.oecd.org/content/dam/oecd/en/publications/reports/2024/04/artificial-intelligence-and-the-changing-demand-for-skills-in-the-labour-market_861a23ea/88684e36-en.pdf
• OECD. “Generative AI and the Future of Work: Global Dialogue” (January 2025).
• Haskel, Jonathan and Stian Westlake. Capitalism Without Capital (2017). Princeton University Press.
• Roughgarden, Tim. “Transaction Fee Mechanism Design: Economic Analysis of EIP-1559.” ACM EC 2021.
• Huberman, Gur, Jacob Leshno, and Ciamac Moallemi. “Monopoly Without a Monopolist.” Review of Economic Studies 88.6 (2021).
• Network Law Review. “Principal-Agent Dynamics in the Age of Agentic AI.” (2025). https://www.networklawreview.org/stocker-lehr-ai/
• California Management Review. “Rethinking AI Agents: A Principal-Agent Perspective.” (2025). https://cmr.berkeley.edu/2025/07/rethinking-ai-agents-a-principal-agent-perspective/
• LSE Centre for Economic Performance. “Technology Adoption and Diffusion Research.” https://cep.lse.ac.uk/_new/OUR-WORK/Growth/New-Technologies-and-Productivity/
• UK CMA. “Foundation Models: Initial Report.” (2024).
• Romer, Paul M. “Endogenous Technological Change.” Journal of Political Economy 98.5 (1990).
• Aghion, Philippe, Benjamin Jones, Charles Jones. “Artificial Intelligence and Economic Growth.” NBER Working Paper No. 23928 (2017).
• Brynjolfsson, Erik et al. “GDP-B: Accounting for the Value of New and Free Goods in the Digital Economy.” NBER Working Paper No. 25695 (2019).
• Autor, David, Dorn, Katz, Patterson, Van Reenen. “The Fall of the Labor Share and the Rise of Superstar Firms.” QJE 135.2 (2020).
• Agrawal, Ajay, Joshua Gans, Avi Goldfarb. Prediction Machines. Harvard Business Review Press (2018).
• Katz, Michael L. and Carl Shapiro. “Network Externalities, Competition, and Compatibility.” AER 75.3 (1985).
• Fehr, Ernst and Simon Gächter. “Cooperation and Punishment in Public Goods Experiments.” AER 90.4 (2000).
• Stigler, George J. “The Theory of Economic Regulation.” Bell Journal of Economics and Management Science 2.1 (1971): 3-21.
• Kahneman, Daniel and Amos Tversky. “Prospect Theory: An Analysis of Decision under Risk.” Econometrica 47.2 (1979): 263-291.
• Akerlof, George A. “The Market for ‘Lemons’: Quality Uncertainty and the Market Mechanism.” Quarterly Journal of Economics 84.3 (1970): 488-500.
• Spence, A. Michael. “Job Market Signaling.” Quarterly Journal of Economics 87.3 (1973): 355-374.
• Roughgarden, Tim. “Transaction Fee Mechanism Design: Economic Analysis of EIP-1559.” ACM EC (2021).
• Fortune. “AI Productivity Paradox CEO Study.” February 2026. https://fortune.com/2026/02/17/ai-productivity-paradox-ceo-study-robert-solow-information-technology-age/
• Williamson, Oliver E. “Transaction-Cost Economics: The Governance of Contractual Relations.” Journal of Law and Economics 22.2 (1979).
• Bresnahan, Timothy F. and Manuel Trajtenberg. “General Purpose Technologies: ‘Engines of Growth’?” Journal of Econometrics 65.1 (1995).
• Olson, Mancur. The Logic of Collective Action. Harvard University Press, 1965.
• UK CMA. “AI Foundation Models: Final Report.” (2024). https://www.gov.uk/government/publications/ai-foundation-models-initial-report
• SiliconANGLE. “Why AI productivity still lags behind investment.” October 2025. https://siliconangle.com/2025/10/02/ai-productivity-ai-hidden-value-uipathfusion/
• Goldman Sachs. “Goldman finds no meaningful relationship between AI and productivity at the economy-wide level.” Yahoo Finance, 2025. https://finance.yahoo.com/news/goldman-finds-no-meaningful-relationship-143553714.html
• Goldman Sachs. “How Will AI Affect the US Labor Market?” Goldman Sachs Insights, 2025. https://www.goldmansachs.com/insights/articles/how-will-ai-affect-the-us-labor-market
• Network Law Review. “Principal-Agent Dynamics and Digital Platform Economics in the Age of Agentic AI.” (2025). https://www.networklawreview.org/stocker-lehr-ai/
• California Management Review. “Rethinking AI Agents: A Principal-Agent Perspective.” (2025). https://cmr.berkeley.edu/2025/07/rethinking-ai-agents-a-principal-agent-perspective/
• Flashbots. “MEV and Me.” (2021-2025). https://writings.flashbots.net — MEV economics and auction mechanism design on Ethereum.
• Census.gov. “Microfoundations of the Productivity J-curves.” CES Working Paper 25-27 (2025). https://www2.census.gov/library/working-papers/2025/adrm/ces/CES-WP-25-27.pdf
• domain-correction: infrastructure → artificial-intelligence. Original stub assigned domain infrastructure which is incorrect; Economics as applied to AI, blockchain, and digital systems is an artificial-intelligence and cross-domain concept, not an infrastructure layer concept. IRI, URI, same-as, legacy-term-id, and owl-class updated accordingly.

Metadata

• domain-correction: infrastructure → artificial-intelligence
• key-theorists: Acemoglu, Restrepo, Brynjolfsson, Rock, Syverson, Tirole, Hart, Holmström, Weyl, Buterin, Coase, Romer, Aghion, Autor, Haskel, Kahneman
• empirical-data-points: Acemoglu-TFP-<0.66pct, Goldman-7pct-GDP, IMF-TFP-1.8pct-5yr, Brynjolfsson-US-productivity-2.7pct-2024, Gitcoin-CO-QF-2.9M, TerraUSD-60B-collapse, Uniswap-AMM-constant-product