Pith. sign in

REVIEW 4 major objections 5 minor 4 cited by

The Agentic Economy

T0 review · 4 major / 5 minor · reviewed 2026-08-07 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read The paper argues that the most disruptive economic impact of generative AI will come from reducing communication frictions between consumers and businesses through AI assistants and service agents that interact programmatically, not from…

desk verdict A clear, honest essay on how AI agents might reshape markets, but the central claim overreaches its own concessions about discovery and trust. read the letter →

arxiv 2505.15799 v2 pith:P46K62EH submitted 2025-05-21 cs.CY

David M. Rothschild , Markus Mobius , Jake M. Hofman , Eleanor W. Dillon , Daniel G. Goldstein , Nicole Immorlica , Sonia Jaffe , Brendan Lucier
show 2 more authors
Aleksandrs Slivkins Matthew Vogel
This is my paper · ORCID
classification cs.CY
keywords generativeAIagentsagenticeconomytwo-sidedplatformscommunicationfrictionsswitchingcostsmicro-transactionspreference
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This paper argues that the most disruptive economic impact of generative AI will come not from making individual humans more productive, but from letting AI agents communicate directly with one another. It imagines an 'agentic economy' in which each consumer has an assistant agent, each business has a service agent, and the two exchange information and negotiate transactions programmatically. Lowering those communication frictions would reduce the cost of switching providers and could make today's two-sided platforms, such as travel, retail, and music intermediaries, less necessary. The authors propose that the architecture of agent communication, whether open like the web or closed like app stores, will determine who captures the gains. This matters because the choices about protocols, standards, and market structure are being made now, before the balance of power is settled.

What carries the argument

The mechanism carrying the argument is the pair of assistant agents and service agents that communicate programmatically in unscripted ways: assistant agents act for consumers, service agents act for businesses, and they negotiate and transact directly with each other rather than through fixed web forms or human browsing. The paper's central analytical tool is the distinction between 'unscripted' interaction, which is technically feasible through natural-language and protocol advances, and 'unrestricted' interaction, which is governed by market power, standards, and regulation. This distinction allows the paper to separate technical capability from economic constraint, and to frame the future as a contest between closed 'agentic walled gardens' and an open 'web of agents.'

What would settle it

If, after several years, leading consumer assistant products still cannot transact directly with service agents operated by other firms, and agent-to-agent transactions remain a negligible share of commerce completed without a human in the loop, the predicted reorganization of markets will not have occurred; one could check this by tracking whether an open agent-to-agent transaction protocol is implemented across major platforms and whether meaningful transaction volume moves through it.

Watch

Extended reading notes

Core claim

The central claim is that unscripted, programmatic communication between assistant agents and service agents will reorganize markets by reducing communication frictions between consumers and businesses. The paper draws a key distinction between 'unscripted' interactions, which are made possible by advances in natural language and protocol design, and 'unrestricted' interactions, which depend on market structure and governance. On this basis it argues that when agent-to-agent communication is cheap and flexible, switching costs fall, the matching and standardization role of two-sided platforms becomes less essential, and new forms of commerce arise: ranking by high-quality consumer feedback rather than attention, micro-transactions instead of subscriptions, and dynamic unbundling and rebundling of digital content. Whether these outcomes materialize depends on whether agents interoperate openly across the economy or are confined to walled gardens controlled by a few dominant providers. The paper's ultimate claim is that this architecture will determine the extent to which generative AI democratizes access to economic opportunity.

Load-bearing premise

The claim depends on consumers and businesses adopting assistant and service agents at scale and on those agents being able to interoperate programmatically across firms, a premise the paper asserts rather than demonstrates.

Editorial extensions

If this is right

  • If agent-to-agent communication becomes cheap and flexible, two-sided platforms will face pressure as intermediaries, although they may retain value through discovery, validation, dispute resolution, and regulatory compliance.
  • The contest between walled gardens and an open web of agents will determine whether market power concentrates in a few firms or spreads across the economy, because open interoperation requires shared standards, trust, and security mechanisms.
  • Advertising and discovery will shift from an attention economy to a preference economy, in which scarce high-quality human feedback becomes the main input for ranking and monetizing service agents.
  • Agent-handled payments make micro-transactions economically viable, favoring usage-based pricing and one-off transactions over long-term subscriptions and repeat business.
  • Digital goods will be unbundled and dynamically rebundled by agents, and with micro-payments in place, a compensation-based retrieval-augmented generation ecosystem can produce personalized content while paying creators.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • If the paper is right, competition policy should focus on interoperability standards and access to agent communication protocols, since control over that layer will become the new bottleneck for market power.
  • A testable extension would run controlled comparisons of consumer choice and prices under agent-assisted switching versus current platform search, measuring whether the assumed switching-cost reduction actually materializes.
  • The walled-garden/open-web framing suggests that even a formally open protocol could still concentrate discovery in dominant providers, so 'unrestricted' communication alone would not guarantee democratization; the paper's own discussion of discovery layers leaves this tension implicit.
  • One can watch for a measurable shift toward variable pricing of digital components, such as per-article or per-feature charges negotiated by agents, as a concrete leading indicator of the micro-transaction prediction.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

4 major / 5 minor

Summary. The paper is a conceptual essay arguing that the most economically transformative impact of generative AI will come from reducing communication frictions between consumers and businesses. It introduces a framework centered on assistant agents and service agents that interact programmatically, and draws a key distinction between unscripted interactions (technically enabled) and unrestricted interactions (governed by market structure). It surveys current siloed and end-to-end agents, and sketches two future scenarios: closed 'agentic walled gardens' and an open 'web of agents.' It then discusses implications for advertising, micro-transactions, and unbundling of digital goods, concluding that the architecture of agentic communication will determine the extent to which generative AI democratizes economic opportunity.

Significance. If the central argument holds, the paper provides a valuable framing for an important emerging question: whether AI agents will reorganize digital markets and weaken the position of two-sided platforms. Its main contribution is a taxonomy and scenario analysis rather than empirical evidence or formal theory. The paper is transparently an essay: it presents no data, no formal model, and no falsifiable predictions. Its strengths are that it identifies a real design tension (unscripted vs. unrestricted communication; walled gardens vs. open web) and connects the argument to ongoing technical standards efforts such as AutoGen, MCP, and Agent2Agent. As a perspective piece, it has clear value, but the strong causal claims in the abstract currently exceed what the argument supports.

major comments (4)
  1. [Section 3.1] The paper's load-bearing claim that 'once communication frictions are low enough, interoperable AI agents could eliminate the necessity for two-sided platforms' is directly undercut by the same section's concession that platforms provide value 'via discovery, validation, remediation, and economies of scale.' Lower communication costs address only one component of intermediation; they do not by themselves solve the discovery problem, the trust problem, or the dispute-resolution problem. If these functions remain scarce and valuable, then the bottleneck is not communication architecture but the governance of the discovery and ranking layer. The authors should either soften the causal claim or provide a concrete mechanism explaining how these platform functions can be unbundled and redistributed across a web of agents without reconcentrating power in a new intermediary layer.
  2. [Section 3.3] The argument that advertising will decline because 'attention is a less constrained resource' is weakened by the same section's statements that 'what matters more is the algorithm that matches assistants to service agents' and that 'the truly scarce and valuable resource... will be high-quality human feedback.' These concessions show that the scarce resource in an agentic economy is not communication bandwidth but the ranking, matching, and feedback-aggregation layer. The paper should reconcile these statements with the abstract's conclusion that 'the architecture of agentic communication will determine' the democratization outcome. As written, the communication architecture is presented as the determining factor even though the paper's own analysis identifies attention and feedback governance as at least equally important.
  3. [Section 3.2] The 'web of agents' scenario is asserted to require 'large-scale coordination among many players' to develop standards and protocols, and 'robust mechanisms for discovery, trust, and security.' The paper does not explain how these coordination problems are solved in the absence of a central intermediary. Since the functions that platforms currently perform would need to be recreated somewhere, the open-web scenario may simply move concentration from transaction platforms to an infrastructure or discovery layer, rather than democratizing access. This is a central weakness because the paper's democratization claim depends on the feasibility of the open web of agents. The authors should either propose a governance mechanism for the discovery layer or explicitly acknowledge that the web-of-agents scenario carries its own centralizing tendencies.
  4. [Abstract and Sections 1-3] The paper makes strong predictive statements, such as 'the architecture of agentic communication will determine the extent to which generative AI democratizes access to economic opportunity,' without offering testable implications or empirical evidence. Since the essay explicitly relies on future technological adoption and interoperability assumptions, the causal claims are not currently falsifiable. The authors should either add a set of specific, observable predictions that could distinguish the walled-garden scenario from the web-of-agents scenario, or reframe the paper explicitly as a speculative scenario analysis rather than a claim about what 'will' happen. This would make the contribution more rigorous and the central claims more honest.
minor comments (5)
  1. [Section 3.1, paragraph 2] The phrase 'eliminate the necessity for two-sided platforms as intermediaries all together' contains a typo: 'all together' should be 'altogether.'
  2. [Section 3.5, final paragraph] The sentence 'We expect more extreme unbundling, rebundling, and new products in a webs of agents where there can be unrestricted communication between the assistant any any service agent' contains two typos: 'webs' should be 'web,' and 'any any' should be 'and any.'
  3. [Section 3.2, closing paragraph] The phrase 'whether walled gardens of web of agents scenario triumphs' should read 'whether the walled-garden or web-of-agents scenario triumphs.'
  4. [References] Reference [2] is a self-citation by two of the authors. It is used only to support the uncontroversial claim that generative AI improves productivity, so it is not load-bearing; nevertheless, the authors should verify that the citation is appropriate and not inflated.
  5. [General] The paper would benefit from a single figure or table contrasting siloed agents, end-to-end agents, agentic walled gardens, and the web of agents along dimensions such as interoperability, governance, and control. This would make the taxonomy easier for readers to apply.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the paper is a speculative framework essay with no fitted parameters, equations, or derivation chains, and its central thesis is argued rather than derived from its own inputs.

full rationale

This paper does not derive a formal result, fit any parameter, or construct a model whose output is encoded in its assumptions. The central claim, that the architecture of agentic communication will determine the degree of economic democratization, is a forward-looking thesis supported by external citations to economic literature (Klemperer on switching costs, Rysman on two-sided markets, Bresnahan and Trajtenberg on general-purpose technologies) and technical standards (AutoGen, MCP, Agent2Agent). The only citation with author overlap is reference [2] (Cui et al., coauthored by Sonia Jaffe), which is used solely as an example that delegation to AI improves individual productivity in the introduction; it is not load-bearing for the paper's central argument about inter-agent communication architecture. No equation or fitted quantity is later relabeled as a prediction, and no uniqueness theorem or prior result by the same authors is invoked to force a conclusion. The paper also explicitly concedes countervailing factors in Section 3.1, acknowledging that intermediaries provide discovery, validation, remediation, and economies of scale, which weakens the strength of its speculative scenario but does not create a circular dependence. The self-referential elements present are ordinary citations to supporting work, not reductions of the thesis to its own premises.

Assumptions & free parameters 0 free parameters · 5 assumptions · 5 invented entities

The paper introduces no free numerical parameters. Its conceptual claims rest on several domain assumptions about adoption, interoperability, communication costs, market structure, and micro-transactions, all stated as likely or necessary rather than demonstrated. The invented entities are scenario constructs and role definitions, not empirically verified objects.

assumptions (5)
  • domain assumption Assistant and service agents will be widely adopted and used for transactions.
    The market reorganization scenarios require broad adoption, which the paper says is 'already underway' in Section 1 but assumes for the future impact.
  • domain assumption Agents will be able to communicate programmatically in unscripted ways at scale.
    Sections 1 and 3.2 assume technical standardization and interoperability across firms; this is plausible but not established.
  • domain assumption Communication frictions are a primary barrier to market efficiency and switching.
    The central thesis in Section 1 relies on communication costs being the key friction; this is asserted rather than measured.
  • domain assumption Market structure, not just technical capability, will determine whether agent interactions are restricted.
    Section 3.2 frames walled gardens versus an open web of agents as driven by market forces and governance, a modeling assumption that is plausible but unverified.
  • domain assumption Micro-transactions will become feasible when handled by agents.
    Section 3.4 assumes that agent handling removes the hassle cost of micro-payments; this is a testable but untested claim.
invented entities (5)
  • assistant agent
    purpose: Act on behalf of consumers in transactions with businesses.
    Partially realized in products like Microsoft Copilot, but the fully interoperable consumer-side agent is hypothetical and has no falsifiable handle in this paper.
  • service agent
    purpose: Represent businesses in programmatic interactions with consumer agents.
    Some siloed examples exist, such as Meta business agents, but the interoperable service agent is assumed as a future construct.
  • agentic walled garden
    purpose: A closed ecosystem where agents can only interact within a platform's domain.
    A scenario label with no independent empirical evidence provided.
  • web of agents
    purpose: An open network where any agent can transact with any other.
    A scenario label with no independent empirical evidence provided.
  • preference economy
    purpose: A hypothesized economy where high-quality human feedback, not attention, is the scarce resource.
    A conceptual construct; no measurement or test is proposed.

how reviews work

0 comments
Cite this review

Pith. "Pith review of The Agentic Economy." pith.science (2026). https://pith.science/paper/P46K62EH

@misc{pith2026250515799,
  author       = {Pith},
  title        = {Pith review of: The Agentic Economy},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/P46K62EH}},
  note         = {Machine review of arXiv:2505.15799}
}
read the original abstract

Generative AI has transformed human-computer interaction by enabling natural language interfaces and the emergence of autonomous agents capable of acting on users' behalf. While early applications have improved individual productivity, these gains have largely been confined to predefined tasks within existing workflows. We argue that the more profound economic impact lies in reducing communication frictions between consumers and businesses. This shift could reorganize markets, redistribute power, and catalyze the creation of new products and services. We explore the implications of an agentic economy, where assistant agents act on behalf of consumers and service agents represent businesses, interacting programmatically to facilitate transactions. A key distinction we draw is between unscripted interactions -- enabled by technical advances in natural language and protocol design -- and unrestricted interactions, which depend on market structures and governance. We examine the current limitations of siloed and end-to-end agents, and explore future scenarios shaped by technical standards and market dynamics. These include the potential tension between agentic walled gardens and an open web of agents, implications for advertising and discovery, the evolution of micro-transactions, and the unbundling and rebundling of digital goods. Ultimately, we argue that the architecture of agentic communication will determine the extent to which generative AI democratizes access to economic opportunity.

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 4 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. When HTTP 402 Meets the Blockchain: Risks on Emerging x402 Payments

    cs.CR 2026-07 conditional novelty 7.0 of 10

    All 15 tested x402 payment facilitators violate at least one of eight derived security rules, yielding four new attack classes and 31 reported vulnerabilities.

  2. From Product Search to Preference Articulation: The Economics of Agentic Commerce

    econ.TH 2026-08 conditional novelty 6.0 of 10

    Manual search collapses as preference complexity grows, while agentic search with consumer refinement stays viable, creates an adoption lag, and can lead platforms to assign lower AI fidelity to attention-rich consumers.

  3. Choose Your Agent: Tradeoffs in Adopting AI Advisors, Coaches, and Delegates in Multi-Party Negotiation

    cs.GT 2026-02 conditional novelty 6.0 of 10

    Users prefer an AI Advisor but gain most with a Delegate, because human editing filters out the AI's best proposals.

  4. Agentic Web: Weaving the Next Web with AI Agents

    cs.AI 2025-07 conditional novelty 3.0 of 10

    A position paper defines the Agentic Web as the next web era and proposes a three-dimensional conceptual framework for understanding and building it.

Reference graph

Works this paper leans on

12 extracted references · 11 canonical work pages · cited by 4 Pith papers

  1. [1]

    Generative ai at work

    Erik Brynjolfsson, Danielle Li, and Lindsey Raymond. Generative ai at work. The Quarterly Journal of Economics , page qjae044, 2025

    work page 2025

  2. [2]

    The effects of generative ai on high skilled work: Evidence from three field experiments with software developers

    Zheyuan Kevin Cui, Mert Demirer, Sonia Jaffe, Leon Musolff, Sida Peng, and Tobias Salz. The effects of generative ai on high skilled work: Evidence from three field experiments with software developers. A vailable at SSRN 4945566, 2024

    work page 2024

  3. [3]

    Competition when consumers have switching costs: An overview with applications to industrial organization, macroeconomics, and international trade

    Paul Klemperer. Competition when consumers have switching costs: An overview with applications to industrial organization, macroeconomics, and international trade. The review of economic studies , 62(4):515–539, 1995

    work page 1995

  4. [4]

    The rapid adoption of generative ai

    Alexander Bick, Adam Blandin, and David J Deming. The rapid adoption of generative ai. Technical report, National Bureau of Economic Research, 2024

    work page 2024

  5. [5]

    The adoption of chatgpt

    Anders Humlum and Emilie Vestergaard. The adoption of chatgpt. Technical report, IZA Discussion Papers, 2024

    work page 2024

  6. [6]

    Autogen, 2024

    Microsoft. Autogen, 2024

    work page 2024

  7. [7]

    Model context protocol (mcp), 2024

    Anthropic PBC. Model context protocol (mcp), 2024

    work page 2024

  8. [8]

    Google agent2agent, 2025

    Google. Google agent2agent, 2025

    work page 2025

Show all 12 references
  1. [9]

    Bresnahan and M

    Timothy F. Bresnahan and M. Trajtenberg. General purpose technologies ‘engines of growth’? Journal of Econometrics, 65(1):83–108, 1995

    1995

  2. [10]

    The economics of two-sided markets

    Marc Rysman. The economics of two-sided markets. Journal of economic perspectives , 23(3):125–143, 2009

    2009

  3. [11]

    Evaluating language model agency through negotiations

    Tim R Davidson, Veniamin Veselovsky, Martin Josifoski, Maxime Peyrard, Antoine Bosselut, Michal Kosinski, and Robert West. Evaluating language model agency through negotiations. arXiv preprint arXiv:2401.04536, 2024

    2024

  4. [12]

    Becker and Kevin M

    Gary S. Becker and Kevin M. Murphy. The division of labor, coordination costs, and knowledge*.The Quarterly Journal of Economics, 107(4):1137–1160, 11 1992

    1992

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.