Orchestrating AI makes it cheaper to modularize systems, but tasks requiring verifiable, auditable, or assignable outputs resist unbundling; firms will either keep capabilities integrated, run parallel tracks, or accept mounting governance 'rule debt'.

Main Finding

Agentic AI orchestrators can make technical interfaces highly modular and dramatically reduce assembly costs, but organizational boundaries do not automatically follow. Whether AI-enabled capabilities become traded components or remain integrated within firms depends on accountability assets (resources that make outputs defensible, auditable, and assignable) and on two key dimensions: verification cost and responsibility transferability. High verification costs or low transferability preserve integrated accountability boundaries; low verification costs and high transferability enable componentization; mixed cases yield dual-track strategies. The theory also identifies rule debt—the latent governance burden that accrues when decision rules migrate out of governed information systems into ungoverned agentic execution environments.

Key Points

• Agentic orchestrators: systems that interpret user intent, select and route tools, monitor intermediate outputs, and autonomously assemble workflows across organizational boundaries. They reduce interface, discovery, and assembly costs, strengthening pressure toward modular, ecosystem-based organization.

• Accountability boundary vs. interface boundary:

  • Interface boundary: where user interaction, invocation, and technical modularity occur.
  • Accountability boundary: where responsibility, evidence, review, signoff, and defensibility are held. These can (and often do) diverge under agentic orchestration.

• Accountability assets: a class of complementary assets whose value derives from making AI outputs legitimate, auditable, reviewable, and assignable. Examples include audit trails, certification processes, professional signoff capacity, evidentiary documentation, and compliance infrastructure. When cospecialized with a capability, they anchor the accountability boundary even if the technical interface is callable.

• Rule debt: governance burden created when decision rules historically embedded in governed IS migrate into agentic prompts or informal instructions. Rule debt requires later inventorying, ownership assignment, versioning, testing, monitoring for drift, and reconciliation with formal policy—costs often unaccounted for in sourcing decisions.

• Two structural dimensions determine boundary choice:

  • Verification cost: the cost of determining whether an AI-supported output is correct, complete, defensible, and usable for organizational action.
  • Responsibility transferability: the ease with which responsibility for an output can be contractually, professionally, or operationally shifted across organizational boundaries.

• Boundary strategies:

  • Component strategy: when verification cost is low and responsibility is transferable—capability is exposed as a callable component.
  • Integrated strategy: when verification cost is high or responsibility is hard to transfer—accountability assets and control retained.
  • Dual-track strategy: when edge functions are componentizable but the accountability-bearing core is retained integrated.

• Platform governance and envelopment: open technical interfaces are not the same as economic openness. Orchestrators control routing, ranking, substitution, and learning; complementors that also surrender accountability assets are more vulnerable to envelopment. Accountability-asset cospecialization reduces vulnerability.

• Seven formal propositions (paper): link agentic reductions in interface/assembly costs, verification cost, responsibility transferability, and accountability-asset cospecialization to the three boundary strategies; elaborate how appropriability and orchestrator intent capture affect value capture; show how boundary misconfiguration creates rule debt.

• Illustrative domains: document processing, legal services, audit, clinical decision support, procurement—show how the accountability logic plays out differently across sectors.

Data & Methods

• Methodological approach: theory development (integrative conceptual synthesis). The authors combine and extend literatures in digital innovation, transaction cost economics, complementary assets/appropriability, digital platform governance, and IS control to build a mid-range theory of agentic accountability boundaries.
• Elements of the method:
  • Concept formation (accountability assets; rule debt).
  • Mechanism specification (how verification costs and responsibility transferability interact with agentic orchestration).
  • Propositional development (seven propositions with derivations, mechanisms, alternative predictions, and scope conditions).
  • Structured theoretical illustrations across multiple domains to discipline and exemplify the theory.
  • A theoretical model (presented in the paper) synthesizing the constructs and causal logic.
• No empirical dataset or formal econometric estimation is reported; scope is conceptual/theoretical with targeted domain examples.

Implications for AI Economics

• Reassessment of Coasian predictions: Lowered assembly and interface costs from agentic AI do not guarantee vertical disintegration. Economists and strategists should incorporate accountability assets and verification costs into make-or-buy and boundary analyses.

• Market structure and firm boundaries:

  • Capabilities that require defensibility (high verification cost or low transferability) will favor integrated provision, supporting vertically integrated firms or specialist accountability-asset holders.
  • Capabilities with low verification cost and transferable responsibility will be commoditized and traded as components, favoring disaggregation and multi-vendor ecosystems.

• Appropriation and rents:

  • When appropriability of core AI functionality is weak (e.g., general-purpose models are broadly available), value capture shifts to owners of accountability assets and to orchestrators that can capture intent and routing rents.
  • Orchestrator intent capture (ability to shape/observe user intent and routing choices) becomes an important channel for appropriation even with open interfaces.

• Platform competition and envelopment:

  • Open APIs are insufficient protection for complementors—control over accountability-related infrastructure is a separate, powerful barrier to substitution/envelopment.
  • Firms that retain or build accountability assets can resist platform substitution; conversely, orchestrators that internalize those assets can more fully enclose ecosystems.

• Hidden transaction and governance costs:

  • Rule debt represents a non-trivial, often unpriced governance liability for organizations using agentic orchestration. This raises the effective transaction costs of outsourcing certain decision functions and may tilt sourcing back toward integration.
  • Markets may emerge for third-party accountability services: verifiers, auditors, evidence-managed APIs, certification providers, and governance middleware that inventory and reconcile agent-executed rules.

• Labor and specialization effects:

  • Demand will rise for professionals and firms that supply accountability assets (audit-capable processes, legally defensible signoff, regulated clinical governance). These roles could preserve rents for incumbents even as front-end tasks are automated.
  • Conversely, commoditization of low-verification tasks will shift complementary labor toward monitoring, evidentiary work, and exception handling.

• Policy and regulation:

  • Standardizing evidentiary and audit requirements (reducing verification costs) or clarifying responsibility transfer rules (improving transferability) would make componentization more feasible and lower economic frictions.
  • Regulators should be aware that agentic orchestration can redistribute accountability in opaque ways and create rule debt that undermines compliance unless addressed.

• Empirical research directions:

  • Measure accountability-asset cospecialization across industries and quantify its impact on boundary choices and market concentration.
  • Quantify rule debt: methods to estimate inventorying/testing costs and their impact on sourcing decisions.
  • Study how orchestrator learning/intent capture translates into rent extraction across ecosystems.

Overall, the paper reframes economic analysis of AI-driven modularization by adding accountability-related complementary assets and governance frictions to the standard cost-and-technology calculus, with important consequences for firm boundaries, platform strategy, appropriation, and regulatory policy.