PreprintCited by 0Digital Platforms and EconomicsTechnology's Impact on MediaArtificial Intelligence Applications

Going Headless? On the Boundaries of Vertical AI Firms

Open access
Hydari, Muhammad Zia, Muzaffar, Farooq·2026·arXiv

Abstract

Vertical AI firms in accounting, law, healthcare, procurement, and similar domains historically bundled workflow, domain logic, and accountability into a single application. General-purpose AI agents are now unbundling that package, prompting founders and investors to advocate "going headless": cede the workflow and interface to agents and expose domain expertise as callable services. This article argues that going headless is correct for some firms and destructive for others, and that the latter often cede their value capture inadvertently through architectural choices that look like interface decisions. This is a boundary question, and the answer turns on distinguishing the interface boundary, which can often move, from the accountability boundary, which often must not. Drawing on Coase's theory of the firm, Eisenmann, Parker, and Van Alstyne's platform envelopment framework, and Teece's analysis of complementary assets and appropriability, the article shows that orchestrators operating through open protocols acquire envelopment power even as technical interoperability improves, and that durable value capture concentrates in cospecialized accountability assets: professional signoff, regulated workflows, evidence trails, and trusted systems of record. The article proposes a three-position taxonomy (component, integrated software platform, dual-track) determined not by sector but by task-accountability regime, and formalizes the construct of rule debt: the future governance, maintenance, and accountability burden that accrues to customer organizations when business rules and professional standards migrate from governed systems into prompts and agent instructions. Four principles follow: decompose by accountability not interface, invert the edges while retaining the core, position rule debt as the customer cost the integrated platform prevents, and avoid single-orchestrator dependence.

AI snapshot

Written from the paper’s full text.

TL;DR

The article argues that the “go headless” prescription for vertical AI firms is correct only where verification is cheap and responsibility is transferable; in accountability-bound domains it can dissipate value capture by ceding the interface while accountability assets remain elsewhere. It proposes distinguishing the movable interface boundary from the accountability boundary, offers a three-position taxonomy, and names rule debt as the customer-side governance burden of migrating rules into prompts.

Key findings

  • Going headless is not universally correct: for component-like tasks such as semi-structured extraction, classification, standards retrieval, and first-pass summarization, headless can be distribution; for accountable judgment such as audit conclusions, clinical governance, or maintenance-deferral decisions, the firm should remain an integrated software platform or retain enforceable control over the accountability layer.
  • Open protocols reduce connection friction but do not remove orchestrator power: envelopment, gatekeeping, and knowledge distillation/product-learning from interface control expose vertical firms to decisions they cannot veto; “the interface is portable; the economic position may not be.”
  • Durable value capture concentrates in cospecialized accountability assets—professional signoff, regulated workflows, evidence trails, and trusted systems of record—especially where appropriability around the AI capability is weak.
  • The paper proposes a three-position taxonomy (component, integrated software platform, dual-track) determined by task-accountability regime rather than sector, and formalizes rule debt; four operating principles follow: decompose by accountability, invert the edges while retaining the core, position rule debt as the customer cost the integrated platform prevents, and avoid single-orchestrator dependence.

Methods

  • Conceptual/strategic article; no empirical dataset or sample is reported. It synthesizes Coase’s theory of the firm, Eisenmann/Parker/Van Alstyne’s platform envelopment framework, and Teece’s complementary assets and appropriability analysis to examine vertical AI firm boundaries.
  • It names three orchestrator relationships—open protocols, curated commercial partnerships, and revenue-sharing—and focuses mainly on the open-protocol pattern; Appendix A applies Williamson’s hold-up framework and the Grossman-Hart-Moore investment-incentive argument to the bilateral-contracting subset.
  • Main analysis is theoretical: it develops boundary distinctions (interface, value-capture, accountability), a taxonomy, and operating principles; no quantitative estimation or empirical test is described.

Limitations

  • Stated: the accountability moat is not absolute; accountability can migrate when outputs standardize, verification becomes statistical, regulators accept new control frameworks, and customers accept a different allocation of responsibility. The taxonomy is a diagnostic snapshot, not a prediction.
  • Stated: the main argument applies most directly to the open-protocol pattern; bilateral contracts change the degree of protection but not the underlying boundary logic. Knowledge-distillation risk is limited by enterprise data-use terms but remains operative where protections are absent, and the boundary between operational observation and learning is difficult to verify externally.
  • Evident from the text: no empirical validation, measurement, or sample; claims rest on conceptual synthesis and illustrative examples (e.g., EMI CT scanner, Hyperscience/Rossum, audit standards), so the magnitude and prevalence of rule debt and value migration are not quantified.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

DataCite · 10.48550/arxiv.2605.17812

Cites

10 of 10 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Claims the paper makes

0 of 32 cited sources checked · 0 supported · 0 flagged

  • Findingreasoning

    Building the capability themselves was rarely competitive because most customers lacked the domain expertise, model access, data, and engineering capacity to match specialist providers.

    “The first option was rarely competitive, since most customers lacked the domain expertise, model access, data, and engineering capacity to match specialist providers”
    Quote in the paperSubstantial claim
  • Interpretationreasoning

    The decision about which parts of the bundle a vertical AI firm should keep inside the firm depends on distinguishing the interface boundary, which can often move, from the accountability boundary, which often must not.

    “The strategic question is which parts of the bundle should remain inside the firm, and the answer turns on distinguishing the interface boundary, which can often move, from the accountability boundary, which often must not.”
    • Also stated in: § Abstract · plain
      “This is a boundary question, and the answer turns on distinguishing the interface boundary (which can often move) from the accountability boundary (which often must not).”
    Quote in the paperSubstantial claim
  • Interpretationnone given

    A three-position taxonomy-component, integrated software platform, and dual-track-is determined by task-accountability regime rather than by sector.

    “The article proposes a threeposition taxonomy (component, integrated software platform, dual-track) determined not by sector but by task-accountability regime”
    Quote in the paperSubstantial claim
  • Interpretationreasoning

    These contractual protections limit the knowledge-distillation mechanism for large enterprise deployments.

    “These contractual protections limit the knowledge-distillation mechanism for large enterprise deployments.”
    Quote in the paperSubstantial claim
  • Interpretationreasoning

    AI agents change which side of the coordination-cost comparison is higher.

    “Agents change which side of the comparison is higher.”
    Quote in the paperSubstantial claim
  • May combine two claims. 45
  • Classified as probably not a substantial claim. 32
  • The quote only partly states this claim. 1
  • May combine two claims.

    Findingcitation

    Three patterns of orchestrator and vertical-firm relationships coexist today: open protocols governed by public specifications, developer terms, or standardized access rules rather than negotiated, relationship-specific bilateral contracts; curated commercial partnerships with negotiated contracts but no revenue share; and explicit revenue-sharing arrangements.

    “Three patterns coexist today: open protocols governed by public specifications, developer terms, or standardized access rules rather than negotiated, relationshipspecific bilateral contracts (the long tail of Model Context Protocol servers, Gemini connectors, the open layer of OpenAI's Apps SDK); curated commercial partnerships with negotiated contracts but no revenue share; and explicit revenue-”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [3]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • May combine two claims.

    Findingcitation

    In practice, enterprise customers increasingly negotiate data-use agreements that prohibit the orchestrator from using interaction data for model training, and the major orchestrator providers offer enterprise terms that exclude commercial or API inputs and outputs from training by default.

    “In practice, enterprise customers increasingly negotiate data-use agreements that prohibit the orchestrator from using interaction data for model training, and the major orchestrator providers (including OpenAI, Anthropic, and Google) offer enterprise terms that exclude commercial or API inputs and outputs from training by default.”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [5]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • May combine two claims.

    Premisecitation

    EMI developed the CT scanner breakthrough, but firms such as GE and Siemens were better positioned to scale the business through manufacturing, hospital sales, and service networks.

    “EMI's CT scanner is the classic example: EMI developed the breakthrough, but firms such as GE and Siemens were better positioned to scale the business through manufacturing, hospital sales, and service networks.”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [5]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • May combine two claims.

    Premisecitation

    Generic complementary assets are widely available and confer no particular advantage; anyone can rent them.

    “Generic complementary assets are widely available and confer no particular advantage; anyone can rent them.”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [5]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • Classified as probably not a substantial claim. May combine two claims.

    Premisecitation

    Teece distinguishes three kinds of complementary assets, and the distinction determines who captures the value.

    “Teece distinguishes three kinds of complementary assets, and the distinction is what determines who captures the value.”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [5]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • Classified as probably not a substantial claim.

    Premisecitation

    Cospecialized complementary assets exhibit two-way dependence: the asset depends on the innovation and the innovation depends on the asset, with no easy substitute on either side.

    “Cospecialized complementary assets exhibit two-way dependence: the asset depends on the innovation and the innovation depends on the asset, with no easy substitute on either side.”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [5]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • Classified as probably not a substantial claim.

    Premisecitation

    Activities are internalized when the costs of searching, coordinating, monitoring, and contracting through the market exceed the costs of managerial direction.

    “Activities are internalized when searching, coordinating, monitoring, and contracting through the market cost more than managerial direction.”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [1]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • May combine two claims.

    Interpretationreasoning

    Whether a vertical AI firm should go headless depends on the firm: the strategy is beneficial for some and harmful for others.

    “This article argues that going headless is correct for some firms and destructive for others”
    • Also stated in: § Abstract · plain
      “That response is correct for some firms and destructive for others.”
    Quote in the paperSubstantial claim
  • May combine two claims. The quote only partly states this claim.

    Interpretationreasoning

    The envelopment dynamic maps onto agent ecosystems through three related mechanisms: envelopment proper, platform gatekeeping, and knowledge distillation and product-learning advantages from interface control.

    “The dynamic maps onto agent ecosystems through three related mechanisms.”
    Quote in the paperSubstantial claim
  • May combine two claims.

    Interpretationreasoning

    The breakthrough belonged to EMI; the value belonged to whoever owned the complementary assets.

    “The breakthrough belonged to EMI; the value belonged to whoever owned the complementary assets.”
    Quote in the paperSubstantial claim
  • May combine two claims.

    Interpretationreasoning

    All three mechanisms expose vertical AI firms to orchestrator decisions they cannot veto, and none requires a contract with the complementor; all three require only that the orchestrator owns the user-facing surface.

    “All three mechanisms expose vertical AI firms to orchestrator decisions they cannot veto, and none requires a contract with the complementor; all three require only that the orchestrator owns the user-facing surface.”
    Quote in the paperSubstantial claim
  • Classified as probably not a substantial claim.

    Premisecitation

    Firms exist because using markets is not costless.

    “Coase argued that firms exist because using markets is not costless [1].”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [1]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • Probably not a substantial claim: may not pass S2 (checkable).

    Interpretationreasoning

    Rule debt is the future governance, maintenance, and accountability burden that accrues to customer organizations when business rules and professional standards migrate from governed systems into prompts and agent instructions.

    “formalizes the construct of rule debt: the future governance, maintenance, and accountability burden that accrues to customer organizations when business rules and professional standards migrate from governed systems into prompts and agent instructions.”
    • Also stated in: § Abstract · plain
      “It can also shift a new liability onto the customer organization: rule debt, the future governance burden that accrues when decision rules migrate from governed systems into prompts and agent instructions.”
    Quote in the paperNot a substantial claim
  • Classified as probably not a substantial claim.

    Interpretationreasoning

    The first mechanism is envelopment proper: the orchestrator builds native capabilities that substitute for the vertical tool, absorbing the functionality into its own offering.

    “The first is envelopment proper: the orchestrator builds native capabilities that substitute for the vertical tool, absorbing the functionality into its own offering.”
    Quote in the paperSubstantial claim
  • May combine two claims.

    Interpretationreasoning

    Open protocols such as the Model Context Protocol reduce one form of friction by standardizing how agents connect to external tools and data, but they do not constrain the orchestrator's residual power to absorb, re-rank, substitute, or distill.

    “Open protocols such as the Model Context Protocol reduce one form of friction by standardizing how agents connect to external tools and data [4], but they do not constrain the orchestrator's residual power to absorb, re-rank, substitute, or distill.”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [4]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • Probably not a substantial claim: may not pass S2 (checkable).

    Premisecitation

    Complementary assets are the resources that turn an invention into a deliverable product: manufacturing capacity, distribution channels, service networks, brand, regulatory approvals, customer relationships, and the operational infrastructure that supports continued use.

    “Complementary assets are the resources that turn an invention into a deliverable product: manufacturing capacity, distribution channels, service networks, brand, regulatory approvals, customer relationships, and the operational infrastructure that supports continued use.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [5]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • Classified as probably not a substantial claim.

    Premisecitation

    Envelopment is the absorption of one platform's functionality into another by exploiting shared user bases and complementary capabilities, pushing the absorbed offering toward commoditization or irrelevance.

    “Eisenmann, Parker, and Van Alstyne describe envelopment as the absorption of one platform's functionality into another by exploiting shared user bases and complementary capabilities, pushing the absorbed offering toward commoditization or irrelevance [3].”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [3]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • May combine two claims.

    Interpretationreasoning

    Customers bought bundles, and vertical AI firms grew up around bundling.

    “So customers bought bundles, and vertical AI firms grew up around bundling.”
    Quote in the paperSubstantial claim
  • May combine two claims.

    Interpretationreasoning

    The interface is portable; the economic position may not be.

    “The interface is portable; the economic position may not be.”
    Quote in the paperSubstantial claim
  • May combine two claims.

    Interpretationreasoning

    The cospecialized category is where strategic outcomes diverge, because cospecialization creates mutual dependence that binds value capture to whoever owns the asset.

    “The cospecialized category is where strategic outcomes diverge, because cospecialization creates mutual dependence that binds value capture to whoever owns the asset.”
    Quote in the paperSubstantial claim
  • Probably not a substantial claim: may not pass S2 (checkable), S4 (based).

    Interpretationnone given

    The practical boundary between operational observation of interaction patterns and learning from them is difficult to verify externally even when contractual restrictions exist.

    “The practical boundary between operational observation of interaction patterns and learning from them is also difficult to verify externally even when contractual restrictions exist.”
    Quote in the paperNot a substantial claim
  • Classified as probably not a substantial claim.

    Interpretationreasoning

    The platform actor in this analysis is the orchestrator, not the vertical AI firm: the orchestrator aggregates users, tools, data sources, and third-party capabilities into the surface where work is requested, routed, and completed.

    “The platform actor in this analysis is the orchestrator, not the vertical AI firm: the orchestrator aggregates users, tools, data sources, and third-party capabilities into the surface where work is requested, routed, and completed.”
    Quote in the paperSubstantial claim
  • Probably not a substantial claim: may not pass S2 (checkable).

    Interpretationreasoning

    An invention without its complementary assets is a laboratory result, not a business.

    “An invention without its complementary assets is a laboratory result, not a business.”
    Quote in the paperNot a substantial claim
  • Classified as probably not a substantial claim.

    Premisecitation

    Specialized complementary assets exhibit one-way dependence: the asset depends on the innovation but the innovation could in principle work with substitute assets.

    “Specialized complementary assets exhibit one-way dependence: the asset depends on the innovation but the innovation could in principle work with substitute assets.”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [5]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • May combine two claims.

    Interpretationcitation

    Frameworks built for bilateral contracting fit the curated and revenue-share patterns, while the open-protocol pattern requires platform-economic logic on platform openness and control instead.

    “Frameworks built for bilateral contracting fit the curated and revenue-share patterns; 4 the open-protocol pattern requires platform-economic logic on platform openness and control instead [2].”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [2]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • Probably not a substantial claim: may not pass S2 (checkable), S4 (based).

    Interpretationnone given

    In industrial maintenance, the decision to defer maintenance on a turbine, refinery component, or aircraft subsystem carries safety, operational, and regulatory exposures that must attach to a named accountable party.

    “In industrial maintenance, a system may predict equipment failure, but the decision to defer maintenance on a turbine, refinery component, or aircraft subsystem carries safety, operational, and regulatory exposures that must attach to a named accountable party.”
    Quote in the paperNot a substantial claim
  • Probably not a substantial claim: may not pass S2 (checkable).

    Interpretationreasoning

    Encoding business rules, professional standards, or operating policies in prompts, saved tasks, ad hoc scripts, and agent instructions rather than in governed systems creates future governance, maintenance, and accountability burden.

    “Rule debt is the future governance, maintenance, and accountability burden created when business rules, professional standards, or operating policies are encoded in prompts, saved tasks, ad hoc scripts, and agent instructions rather than in governed systems.”
    Quote in the paperNot a substantial claim
  • May combine two claims.

    Interpretationreasoning

    A firm following the headless prescription in an accountability-bound domain may give away the value-capture surface, and the discovery typically comes only after the migration is complete and the assets have already begun to dissipate.

    “A firm following the headless prescription in an accountability-bound domain may discover that it has not just moved its interface to the orchestrator but given away the value-capture surface, and the discovery typically comes only after the migration is complete and the assets have already begun to dissipate.”
    Quote in the paperNot a substantial claim
  • May combine two claims.

    Interpretationreasoning

    A vertical firm's investments in evidence systems, review processes, and escalation paths are not substitutable across domains: audit evidence infrastructure does not transfer to clinical decision support, and a clinical workflow is useless to an auditor.

    “A vertical firm's investments in evidence systems, review processes, and escalation paths are not substitutable across domains: audit evidence infrastructure does not transfer to clinical decision support, and a clinical workflow is useless to an auditor.”
    Quote in the paperNot a substantial claim
  • Probably not a substantial claim: may not pass S2 (checkable). May combine two claims.

    Interpretationreasoning

    Bilateral contracting changes the degree of protection but not the underlying boundary logic.

    “But bilateral contracting does not change the central prescription of the article. It changes the degree of protection, not the underlying boundary logic.”
    Quote in the paperNot a substantial claim
  • May combine two claims.

    Premisereasoning

    The distinction between component, integrated software platform, and dual-track strategic positions is based on task-accountability regime, not sector, and a single firm may contain all three.

    “This yields three strategic positions: component, integrated software platform, 6 and dual-track.The distinction is not by sector but by task-accountability regime.A healthcare firm, a legal technology firm, or a procurement platform may contain all three at once.”
    Quote in the paperSubstantial claim
  • Probably not a substantial claim: may not pass S4 (based). May combine two claims.

    Interpretationnone given

    Dual-track firms have two businesses: a component business optimizing for adoption, usage, interoperability, latency, and gross margin, and a platform business optimizing for retention, governance, risk reduction, evidence, and trust.

    “Dual-track firms have two businesses, not one.The component business optimizes for adoption, usage, interoperability, latency, and gross margin.The platform business optimizes for retention, governance, risk reduction, evidence, and trust.”
    Quote in the paperNot a substantial claim
  • Classified as probably not a substantial claim.

    Interpretationreasoning

    For vertical AI firms, accountability assets (evidence trails, evaluation infrastructure, professional review processes, escalation paths, regulated workflows, proprietary domain data, customer trust, and systems-of-record integrations) may support an orchestrator partnership but do not depend on any single orchestrator for their value.

    “For vertical AI firms, those assets are usually the accountability assets: evidence trails, evaluation infrastructure, professional review processes, escalation paths, regulated workflows, proprietary domain data, customer trust, and systems-of-record integrations.These assets may support an orchestrator partnership, but they do not depend on any single orchestrator for their value.”
    Quote in the paperSubstantial claim
  • May combine two claims.

    Interpretationreasoning

    Many vertical AI firms were built around owning workflow, and their architectures and sales processes assumed that work would happen inside their product; moving from platform to component can cannibalize the interface, confuse buyers, and expose that the prior moat was orchestration rather than judgment.

    “Many vertical AI firms were built around owning workflow.Their architectures and sales processes assumed that work would happen inside their product.Moving from platform to component is therefore not a simple API launch.It can cannibalize the interface, confuse buyers, and expose that the prior moat was orchestration rather than judgment.”
    Quote in the paperNot a substantial claim
  • May combine two claims.

    Interpretationreasoning

    A bilateral contract gives the vertical firm more protection than an open-protocol participant has, but it does not eliminate the boundary problem.

    “That contract matters.It gives the vertical firm more protection than an open-protocol participant has.But it does not eliminate the boundary problem.”
    Quote in the paperSubstantial claim
    • Rests on an assumption: The orchestrator usually owns the user-facing surface, routing logic, ranking rules, evaluation criteria, and terms under which third-party capabilities are discovered.
  • Probably not a substantial claim: may not pass S2 (checkable), S4 (based).

    Interpretationreasoning

    A headless strategy bound to one agent recreates the envelopment exposure it was meant to avoid.

    “A headless strategy bound to one agent recreates the envelopment exposure it was meant to avoid.”
    Quote in the paperNot a substantial claim
  • May combine two claims.

    Interpretationreasoning

    The AI capability makes these assets more scalable and efficient, while these assets give the AI capability its legitimacy, defensibility, and operational integration.

    “The AI capability makes these assets more scalable and efficient, while these assets give the AI capability its legitimacy, defensibility, and operational integration.”
    Quote in the paperSubstantial claim
  • May combine two claims.

    Interpretationcitation

    The accountability moat is not absolute; some activities once treated as integrated professional judgment have been standardized and unbundled, and accountability can migrate when outputs become standardized, verification becomes statistical, regulators accept new control frameworks, and customers accept a different allocation of responsibility.

    “The accountability moat is not absolute.Some activities once treated as integrated professional judgment have been standardized and unbundled.Technology-assisted review made large-scale legal document review more tool-mediated.Robo-advisers brought algorithmic portfolio management into a regulated advisory setting, while remaining subject to fiduciary obligations as registered investment advisers.”
    Quote in the paperSubstantial claim0 of 2 sources checked
    • cites [9]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • cites [10]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • May combine two claims.

    Interpretationreasoning

    An orchestrator's investments in planning quality, tool routing, and general reasoning capability are highly substitutable across domains: the same planning engine serves accounting, law, healthcare, and many other applications, so the orchestrator amortizes those investments across an enormous customer base and has strong incentive to make them.

    “An orchestrator's investments in planning quality, tool routing, and general reasoning capability are highly substitutable across domains: the same planning engine serves accounting, law, healthcare, and a hundred other applications, so the orchestrator amortizes those investments across an enormous customer base and has strong incentive to make them.”
    Quote in the paperNot a substantial claim
  • Classified as probably not a substantial claim.

    Premisecitation

    Teece's framework adds a second variable that interacts with complementary-asset ownership: appropriability.

    “Teece's framework adds a second variable that interacts with complementary-asset ownership: appropriability.”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [5]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • Probably not a substantial claim: may not pass S2 (checkable). May combine two claims.

    Interpretationreasoning

    Treating the component and platform businesses as one prices accountability like traffic and evaluates governed work like an API call.

    “Treating them as one business prices accountability like traffic and evaluates governed work like an API call.”
    Quote in the paperNot a substantial claim
  • May combine two claims.

    Interpretationreasoning

    Firms invest more in assets whose returns they expect to capture; if a vertical firm's returns depend mainly on the orchestrator's future routing and ranking choices, it has weaker incentives to invest deeply in that relationship, whereas if it owns or controls accountability assets valuable regardless of orchestrator, it has stronger incentives to keep improving them.

    “This matters because firms invest more in assets whose returns they expect to capture.If the vertical firm's returns depend mainly on the orchestrator's future routing and ranking choices, the vertical firm has weaker incentives to invest deeply in that relationship.If, instead, the vertical firm owns or controls the accountability assets that remain valuable regardless of which orchestrator is”
    Quote in the paperSubstantial claim
  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    Audit evidence standards emphasize that reliability depends on the nature and source of evidence and the circumstances under which it is obtained.

    “Audit evidence standards also emphasize that reliability depends on the nature and source of evidence and the circumstances under which it is obtained [6, 7].”
    Quote in the paperNot a substantial claim0 of 2 sources checked
    • cites [6]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • cites [7]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • Probably not a substantial claim: may not pass S2 (checkable), S4 (based). May combine two claims.

    Interpretationreasoning

    Most accountability-bound domains fall into the second category, which is why the vertical firm's boundary should retain ownership of cospecialized assets even when the interface boundary moves.

    “Most accountability-bound domains fall into the second category, which is why the vertical firm's boundary should retain ownership of cospecialized assets even when the interface boundary moves.”
    Quote in the paperNot a substantial claim
    • Rests on an assumption: Most accountability-bound domains fall into the category where domain-specific accountability infrastructure matters most.
  • May combine two claims.

    Premisecitation

    Contracts cannot specify everything that may happen in the future; a partnership contract may say how an integration works today but cannot fully specify what happens if the orchestrator changes its ranking logic, launches a competing native feature, alters its evaluation method, changes its user interface, or extends the protocol in a way that favors some partners over others.

    “The Grossman-Hart-Moore framework starts from another simple observation: contracts cannot specify everything that may happen in the future.A partnership contract may say how an integration works today, but it cannot fully specify what happens if the orchestrator changes its ranking logic, launches a competing native feature, alters its evaluation method, changes its user interface, or extends the”
    Quote in the paperSubstantial claim0 of 2 sources checked
    • cites [14]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • cites [15]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • May combine two claims.

    Premisecitation

    Some investments are much more valuable inside a particular relationship than outside it, and once made, the counterparty may have leverage because the investing firm cannot easily walk away without losing much of the value it has created.

    “Williamson's hold-up framework begins with a simple observation: some investments are much more valuable inside a particular relationship than outside it.Once those investments are made, the counterparty may have leverage, because the investing firm cannot easily walk away without losing much of the value it has created.”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [13]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • May combine two claims.

    Interpretationnone given

    Bilateral contracting improves the vertical firm's position relative to the open-protocol case, allowing a curated partner to negotiate enforceable terms and a revenue-sharing partner to secure commercial participation, and both to obtain more notice, predictability, and formal remedies.

    “Bilateral contracting improves the vertical firm's position relative to the open-protocol case.A curated partner can negotiate enforceable terms.A revenue-sharing partner can secure commercial participation in the orchestrator's success.Both can obtain more notice, more predictability, and more formal remedies than an ordinary open-protocol participant.”
    Quote in the paperNot a substantial claim
  • May combine two claims.

    Premisecitation

    Platform research shows that APIs and developer ecosystems can move value creation outside the firm, but openness also changes control and risk.

    “Platform research shows that APIs and developer ecosystems can move value creation outside the firm, but openness also changes control and risk [11].”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [11]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • Classified as probably not a substantial claim. May combine two claims.

    Interpretationreasoning

    The open-protocol case, the curated-partnership case, and the revenue-sharing case differ in legal structure but converge on the same strategic lesson: the interface can move toward the orchestrator, but the accountability boundary should move only when accountability itself has migrated.

    “The open-protocol case, the curated-partnership case, and the revenue-sharing case therefore differ in legal structure but converge on the same strategic lesson.The interface can move toward the orchestrator.The accountability boundary should move only when accountability itself has migrated.”
    Quote in the paperSubstantial claim
  • Probably not a substantial claim: may not pass S2 (checkable), S4 (based). May combine two claims.

    Interpretationnone given

    The component-tier business does not require foundation models to remain weak at parsing; it requires only that the accuracy gap between general and specialized infrastructure remain meaningful in high-stakes document workflows.

    “The component-tier business does not require foundation models to remain weak at parsing; it requires only that the accuracy gap between general and specialized infrastructure remain meaningful in high-stakes document workflows.”
    Quote in the paperNot a substantial claim
    • Rests on an assumption: The accuracy gap between general and specialized infrastructure remains meaningful in high-stakes document workflows.
  • May combine two claims.

    Interpretationreasoning

    The vertical firm accumulates these domain-specific assets slowly through years of client and regulatory engagement, and only it has reason to invest in them.

    “The vertical firm accumulates these domain-specific assets slowly through years of client and regulatory engagement, and only it has reason to invest in them.”
    Quote in the paperSubstantial claim
  • May combine two claims.

    Interpretationreasoning

    When a firm exposes its capability as a callable tool while ceding the workflow, the orchestrator controls the user relationship, the evaluation surface, the audit trail of which tools were invoked, and the surface where signoff occurs.

    “In practice, the orchestrator now controls the user relationship, the evaluation surface, the audit trail of which tools were invoked, and the surface where signoff occurs.”
    Quote in the paperNot a substantial claim
  • May combine two claims.

    Interpretationreasoning

    In accountability-bound domains, errors are costly, verification is difficult, context accumulates over time, and responsibility cannot be cleanly transferred to a general agent.

    “Here errors are costly, verification is difficult, context accumulates over time, and responsibility cannot be cleanly transferred to a general agent.”
    Quote in the paperNot a substantial claim
  • Probably not a substantial claim: may not pass S4 (based). May combine two claims.

    Interpretationreasoning

    Specialized assets such as proprietary domain data tilt advantage toward whoever owns them, but the AI capability could in principle work with alternative data sources.

    “Specialized assets such as proprietary domain data tilt advantage toward whoever owns them, but the AI capability could in principle work with alternative data sources.”
    Quote in the paperNot a substantial claim
  • Classified as probably not a substantial claim. May combine two claims.

    Interpretationreasoning

    Together they identify three boundaries a vertical AI firm must distinguish: the interface boundary (where the product meets the orchestrator), the value-capture boundary (where appropriability concentrates), and the accountability boundary (where responsibility, evidence, and governance must reside).

    “Together they identify three boundaries a vertical AI firm must distinguish: the interface boundary (where the product meets the orchestrator), the value-capture boundary (where appropriability concentrates), and the accountability boundary (where responsibility, evidence, and governance must reside).”
    Quote in the paperSubstantial claim
  • Probably not a substantial claim: may not pass S4 (based).

    Interpretationnone given

    Customer organizations adopting agentic automation accumulate rule debt when business rules, professional standards, and operating policies migrate into prompts and agent instructions rather than into governed systems.

    “Customer organizations adopting agentic automation accumulate rule debt when business rules, professional standards, and operating policies migrate into prompts and agent instructions rather than into governed systems.”
    Quote in the paperNot a substantial claim
  • Probably not a substantial claim: may not pass S4 (based).

    Interpretationnone given

    In vertical AI, generic assets such as compute confer little advantage because anyone can rent compute.

    “In vertical AI, generic assets such as compute confer little advantage because anyone can rent compute.”
    Quote in the paperNot a substantial claim
  • Probably not a substantial claim: may not pass S2 (checkable).

    Premisecitation

    Appropriability refers to how easily an innovator can prevent others from imitating the innovation itself, separate from any complementary asset.

    “Appropriability refers to how easily an innovator can prevent others from imitating the innovation itself, separate from any complementary asset.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [5]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • Probably not a substantial claim: may not pass S2 (checkable). May combine two claims.

    Premisecitation

    Strong appropriability comes from patents, secrecy, tacit knowledge, or lead-time advantages that competitors cannot quickly close; weak appropriability means the innovation can be freely replicated.

    “Strong appropriability comes from patents, secrecy, tacit knowledge, or lead-time advantages that competitors cannot quickly close; weak appropriability means the innovation can be freely replicated.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [5]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • May combine two claims.

    Interpretationreasoning

    Where appropriability is strong, through proprietary fine-tuning data that competitors cannot replicate, accumulated evaluation sets tightly coupled to the model, or genuinely tacit domain knowledge embedded in deployment, the firm has additional optionality and may capture value without owning the full accountability stack.

    “Where appropriability is strong, through proprietary fine-tuning data that competitors cannot replicate, accumulated evaluation sets tightly coupled to the model, or genuinely tacit domain knowledge embedded in deployment, the firm has additional optionality and may capture value without owning the full accountability stack.”
    Quote in the paperSubstantial claim
  • May combine two claims.

    Interpretationreasoning

    Control over cospecialized assets materially improves a firm's ability to appropriate value, and that is where boundary choices over cospecialized assets become consequential.

    “Control over cospecialized assets materially improves a firm's ability to appropriate value, and that is where boundary choices over cospecialized assets become consequential.”
    Quote in the paperNot a substantial claim
  • Probably not a substantial claim: may not pass S2 (checkable).

    Interpretationreasoning

    A prompt that tells an agent how to classify a transaction, escalate a claim, summarize a medical message, assess a supplier, or approve an exception is an informal policy document.

    “A prompt that tells an agent how to classify a transaction, escalate a claim, summarize a medical message, assess a supplier, or approve an exception is not just an instruction.It is an informal policy document.”
    Quote in the paperNot a substantial claim
  • May combine two claims.

    Interpretationreasoning

    The interface boundary can often move; the accountability boundary typically should not.

    “The first can often move; the third typically should not.”
    Quote in the paperNot a substantial claim
  • May combine two claims.

    Interpretationreasoning

    Where the customer's combined output depends most on domain-specific accountability infrastructure, the vertical firm is the rational investor and owner or controller; under arrangements that leave those assets controlled elsewhere, no party has the incentive to invest in them at the level the domain requires, and the accountability infrastructure deteriorates.

    “Where the customer's combined output depends most on domain-specific accountability infrastructure, the vertical firm is the rational investor and owner or controller; under arrangements that leave those assets controlled elsewhere, no party has the incentive to invest in them at the level the domain requires, and the accountability infrastructure deteriorates.”
    Quote in the paperNot a substantial claim
    • Rests on an assumption: Under arrangements that leave cospecialized assets controlled elsewhere, no party has incentive to invest in them at the level the domain requires.
  • May combine two claims.

    Interpretationreasoning

    When the contract is silent, control over the relevant asset matters; the orchestrator usually controls the interface, marketplace, routing system, user relationship, and rules by which third-party tools are surfaced, while the vertical firm controls or should try to control domain-specific assets that make its outputs trustworthy.

    “When the contract is silent, control over the relevant asset matters.The orchestrator usually controls the interface, the marketplace, the routing system, the user relationship, and the rules by which third-party tools are surfaced.The vertical firm controls, or should try to control, the domain-specific assets that make its outputs trustworthy: the evidence chain, the review workflow, the audit”
    Quote in the paperSubstantial claim
    • Rests on an assumption: The orchestrator usually controls the interface, marketplace, routing system, user relationship, and rules by which third-party tools are surfaced.
  • Probably not a substantial claim: may not pass S2 (checkable), S4 (based). May combine two claims.

    Interpretationreasoning

    The vertical AI firms that survive will make their expertise easy to call and their accountability hard to copy, while losers will celebrate going headless while quietly giving up the assets that made them vertical, and the surrender will show up in underinvestment in the things they no longer own or control.

    “The vertical AI firms that survive will not defend every screen, nor will they dissolve themselves into someone else's agent.They will make their expertise easy to call and their accountability hard to copy.The losers will celebrate going headless while quietly giving up the assets that made them vertical, and the surrender will show up not in their announcements but in their underinvestment in”
    Quote in the paperNot a substantial claim
66 of 66
  • The paper relies on this without support.

    Vertical AI firms historically bundled workflow, domain logic, and accountability into a single application.

    “Vertical AI firms-the companies that have built specialized AI products for accounting, law, healthcare, procurement, software engineering, insurance, and similar domains-have historically bundled several activities into a single application.”
  • The paper relies on this without support.

    General-purpose AI agents are now beginning to unbundle that bundle.

    “General-purpose AI agents are now beginning to unbundle it.”
  • The paper relies on this without support.

    Orchestrators carry distribution to hundreds of millions of users that an individual vertical firm cannot match alone.

    “Orchestrators carry distribution to hundreds of millions of users an individual vertical firm cannot match alone.”
  • The paper relies on this without support.

    Offloading workflow assembly to a general agent saves engineering capacity that can be redirected to the domain capability itself.

    “And offloading workflow assembly to a general agent saves engineering capacity that can be redirected to the domain capability itself.”
  • The paper relies on this without support.

    In some domains, customers buy accountable judgment rather than task completion.

    “In domains where customers buy accountable judgment rather than task completion”
  • The paper relies on this without support.

    Almost every vertical AI firm will integrate with agents.

    “almost every firm will”
  • The paper relies on this without support.

    The headless prescription makes accountability infrastructure look like ordinary workflow.

    “the headless prescription makes accountability infrastructure look like ordinary workflow”
  • The paper relies on this without support.

    Historically, market coordination cost exceeded the premium vertical AI firms charged to deliver the bundle pre-assembled.

    “Historically, market coordination cost was the higher of the two: integrating components, handling errors, reconciling schemas, and maintaining the connective glue typically exceeded the premium vertical AI firms charged to deliver the bundle pre-assembled.”
  • The paper relies on this without support.

    Agents collapse the market coordination cost that previously made assembly uneconomic.

    “the agent collapses the market coordination cost that previously made assembly uneconomic”
  • The paper relies on this without support.

    The open-protocol pattern is the dominant case for the vertical AI firms the article is about.

    “which is the dominant case for the vertical AI firms the article is about”
  • The paper relies on this without support.

    Firms invest in assets whose returns they expect to capture and avoid investing in assets whose returns accrue to others.

    “Firms invest in assets whose returns they expect to capture and avoid investing in assets whose returns will accrue to someone else.”
  • The paper relies on this without support.

    Most accountability-bound domains fall into the category where domain-specific accountability infrastructure matters most.

    “Most accountability-bound domains fall into the second category”
  • The paper relies on this without support.

    Under arrangements that leave cospecialized assets controlled elsewhere, no party has incentive to invest in them at the level the domain requires.

    “under arrangements that leave those assets controlled elsewhere, no party has the incentive to invest in them at the level the domain requires”
  • The paper relies on this without support.

    Customers often prefer capabilities inside existing agents, enterprise workflows, or internal systems rather than standalone applications.

    “The customer often does not want another standalone application.The customer wants the capability available inside an agent, an enterprise workflow, or an internal system.”
  • The paper relies on this without support.

    The accuracy gap between general and specialized infrastructure remains meaningful in high-stakes document workflows.

    “it requires only that the accuracy gap between general and specialized infrastructure remain meaningful in high-stakes document workflows.”
  • The paper relies on this without support.

    The orchestrator usually owns the user-facing surface, routing logic, ranking rules, evaluation criteria, and terms under which third-party capabilities are discovered.

    “The orchestrator still usually owns the user-facing surface, the routing logic, the ranking rules, the evaluation criteria, and the terms under which third-party capabilities are discovered.”
  • The paper relies on this without support.

    The orchestrator usually controls the interface, marketplace, routing system, user relationship, and rules by which third-party tools are surfaced.

    “The orchestrator usually controls the interface, the marketplace, the routing system, the user relationship, and the rules by which third-party tools are surfaced.”
17 of 17

Cites

Flagged first, then by citations

Authors
Hydari, M. Z., & Muzaffar, F.
Affiliations
University of Pittsburgh (US)
Venue
arXiv
Publisher
arXiv
Published
2026
ISSN
2331-8422 (ISSN-L)
OpenAlex
W7161718076
Cited by
0
References
8
Type
posted content
Open access
Locations
  • arXiv (Cornell University) (submittedVersion)
  • arXiv (Cornell University)
Topics
  • Social Sciences › Business, Management and Accounting › Strategy and Management › Digital Platforms and Economics
  • Social Sciences › Decision Sciences › Information Systems and Management › Technology's Impact on Media
  • Physical Sciences › Computer Science › Artificial Intelligence › Artificial Intelligence Applications
Keywords
interface boundaries, platform envelopment, complementary assets, appropriability, AI governance

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

Claims read by deepseek-v4.1-flash under definition version 3 (extract-v3) on 2026-10-08.

Classified by jev-1.13-20260917 (gold-guideline-v0.2) on 2026-10-08.

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review