arxiv: 2604.26482 · v2 · submitted 2026-04-29 · 💻 cs.CY

Recognition: unknown

The Buy-or-Build Decision, Revisited: How Agentic AI Changes the Economics of Enterprise Software

David Klotz

Pith reviewed 2026-05-07 10:45 UTC · model grok-4.3

classification 💻 cs.CY

keywords make-or-buy decisiontransaction cost economicsagentic AIenterprise softwareSaaSgovernance structuresresource-based viewsoftware development economics

0 comments

The pith

Agentic AI transforms the make option from pure hierarchy into a hybrid governance form dependent on external AI infrastructure.

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

The paper re-evaluates the classic make-or-buy decision for enterprise software by combining transaction cost economics with an assessment of current agentic AI coding systems. It shows how AI alters seven key determinants including cost, asset specificity, vendor lock-in, time-to-market, and organizational capability. The central finding is that AI shifts the governance properties of in-house development away from Williamson's pure hierarchy toward a hybrid model that retains code ownership but adds dependency on external AI tools. This challenges the SaaSocalypse narrative that AI will make most SaaS obsolete, instead producing a typology where make is favored only for commodity utilities and differentiating custom applications while regulated and mission-critical systems remain in the buy category.

Core claim

AI fundamentally transforms the governance properties of the Make option, shifting it from Williamson's pure hierarchy to a hybrid governance form that combines code ownership with external AI infrastructure dependency, with qualitatively different economics, capability requirements, and governance structures than pre-AI in-house development.

What carries the argument

The seven canonical decision determinants (cost, strategic differentiation, asset specificity, vendor lock-in, time-to-market, quality and compliance, organizational capability) re-assessed through agentic AI capabilities to produce a typology of enterprise applications by AI sensitivity.

Load-bearing premise

A qualitative assessment of current agentic AI capabilities is sufficient to determine the direction and magnitude of shifts in the seven decision determinants without empirical data or capability benchmarks.

What would settle it

Empirical observation of actual enterprise adoption patterns, measured cost reductions, and governance changes in specific application categories after widespread agentic AI tool deployment.

read the original abstract

Advances in generative artificial intelligence, particularly agentic coding systems capable of autonomous software development, are disrupting the economics of the make-or-buy decision for enterprise applications. The "SaaSocalypse" narrative predicts that AI will render large segments of the Software-as-a-Service market obsolete by enabling firms to build software in-house at a fraction of historical cost. This paper adopts a conceptual research approach, combining transaction cost economics and the resource-based view with an assessment of current AI capabilities, to systematically re-evaluate the factors underlying the make-or-buy decision. It makes three contributions. First, it provides a factor-level analysis of how AI reshapes seven canonical decision determinants: cost, strategic differentiation, asset specificity, vendor lock-in, time-to-market, quality and compliance, and organizational capability. Second, it develops a typology of enterprise applications by their sensitivity to AI-induced shifts in make-or-buy economics. Third, it demonstrates that AI fundamentally transforms the governance properties of the Make option, shifting it from Williamson's pure hierarchy to a hybrid governance form that combines code ownership with external AI infrastructure dependency, with qualitatively different economics, capability requirements, and governance structures than pre-AI in-house development. The analysis finds that the SaaSocalypse thesis is overstated for most enterprise application categories; Make is most compelling for commodity utilities and differentiating custom applications in the AI era, while regulated and mission-critical systems remain predominantly in the buy domain.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

The paper introduces a typology of enterprise apps by AI sensitivity and reframes in-house development as hybrid governance, but its claims about shifts in the seven factors rest on untested qualitative judgments of agentic AI.

read the letter

The paper's core value is the typology that groups enterprise applications by how much agentic AI changes their make-or-buy economics, plus the reframing of the Make option as a hybrid form that mixes code ownership with dependence on external AI infrastructure. It walks through the standard seven determinants from transaction cost economics and the resource-based view and shows why the full SaaSocalypse story does not hold for regulated or mission-critical systems while Make stays attractive for commodity utilities and high-differentiation custom work. That breakdown is clear and applies established theory to the new setting without obvious circularity. The hybrid governance point is a reasonable extension once you accept that ongoing AI tool access creates a persistent external dependency unlike traditional in-house builds. The analysis draws on prior literature in a straightforward way and avoids fitting parameters to its own assertions. The main limitation is that every asserted change in cost, specificity, lock-in, time-to-market, quality, and capability is derived from the author's reading of current AI performance rather than data, benchmarks, case evidence, or even detailed examples of deployed agentic systems. Without criteria for when an external AI dependency becomes governance-altering rather than incremental like cloud services, the direction and size of the shifts remain hard to evaluate. The typology and the rejection of broad SaaS displacement therefore inherit the same ungrounded mapping from capability description to economic outcome. This is the kind of conceptual piece that IT strategy researchers or enterprise architects could use as a starting framework for discussion. It will not satisfy readers who want falsifiable predictions or empirical tests. The work shows clear engagement with the literature and deserves a serious referee to pressure-test the typology and the hybrid claim, even though it will likely need added grounding on AI capabilities before publication.

Referee Report

2 major / 2 minor

Summary. The paper claims that agentic AI systems disrupt the make-or-buy decision for enterprise software by reshaping seven canonical determinants (cost, strategic differentiation, asset specificity, vendor lock-in, time-to-market, quality/compliance, and organizational capability). Drawing on transaction cost economics and the resource-based view, it develops a typology of application categories by sensitivity to these shifts and concludes that the Make option transforms from Williamson's pure hierarchy into a hybrid governance form (code ownership plus external AI infrastructure dependency), rendering the SaaSocalypse narrative overstated; Make is most viable for commodity utilities and differentiating custom apps, while regulated/mission-critical systems remain in the buy domain.

Significance. If the qualitative capability assessments hold, the paper offers a useful conceptual synthesis that integrates established TCE/RBV frameworks with AI-specific factors and supplies a typology that could structure future empirical studies on software sourcing. Its main value lies in the governance reframing of the Make option and the rejection of blanket disruption claims, though this significance is conditional on validation of the unbenchmarked judgments.

major comments (2)

[factor-level analysis] Factor-level analysis of the seven determinants: the asserted directional and qualitative shifts in asset specificity, vendor lock-in, and organizational capability (and thus the hybrid governance characterization) rest entirely on an unbenchmarked qualitative description of current agentic AI capabilities, without performance metrics, adoption data, sensitivity checks, or explicit criteria for when an external AI dependency crosses from incremental (e.g., cloud/compilers) to governance-transforming. This mapping is load-bearing for both the typology and the central claim that Make is no longer a pure hierarchy.
[typology of enterprise applications] Typology of enterprise applications and SaaSocalypse rejection: the classification of categories (commodity utilities, differentiating custom, regulated/mission-critical) and the conclusion that Make is compelling only for the first two inherit the same untested qualitative-to-economic mapping; no case examples, falsifiable thresholds, or robustness checks are supplied to support the differential sensitivity claims.

minor comments (2)

[Abstract] The abstract lists three contributions but does not map them to specific sections or subsections, which would improve navigation in a conceptual paper.
The term 'SaaSocalypse' is introduced in the abstract but would benefit from a brief definitional sentence on first use in the main text to aid readers unfamiliar with the narrative.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive and detailed comments on our conceptual paper. We address each major comment below, clarifying the theoretical basis of our analysis while proposing targeted revisions to enhance transparency.

read point-by-point responses

Referee: [factor-level analysis] Factor-level analysis of the seven determinants: the asserted directional and qualitative shifts in asset specificity, vendor lock-in, and organizational capability (and thus the hybrid governance characterization) rest entirely on an unbenchmarked qualitative description of current agentic AI capabilities, without performance metrics, adoption data, sensitivity checks, or explicit criteria for when an external AI dependency crosses from incremental (e.g., cloud/compilers) to governance-transforming. This mapping is load-bearing for both the typology and the central claim that Make is no longer a pure hierarchy.

Authors: We agree that the analysis relies on a qualitative assessment of current agentic AI capabilities, drawn from the existing literature and observed industry developments rather than new empirical benchmarks. As a conceptual contribution integrating transaction cost economics and the resource-based view, the paper's purpose is to derive theoretical implications for governance structures, not to empirically validate capability levels. The hybrid governance reframing follows directly from the logic that code ownership remains with the firm while development infrastructure becomes dependent on external AI providers, creating a distinct form from both pure hierarchy and market. To address the load-bearing concern, we will add explicit criteria for governance-transforming dependencies (e.g., based on the degree of autonomous code generation and substitutability of the AI service) and include sensitivity discussions for varying AI capability assumptions. revision: partial
Referee: [typology of enterprise applications] Typology of enterprise applications and SaaSocalypse rejection: the classification of categories (commodity utilities, differentiating custom, regulated/mission-critical) and the conclusion that Make is compelling only for the first two inherit the same untested qualitative-to-economic mapping; no case examples, falsifiable thresholds, or robustness checks are supplied to support the differential sensitivity claims.

Authors: The typology is derived conceptually by applying the seven factor shifts to the distinguishing attributes of each application category, such as regulatory constraints that limit AI-driven development in mission-critical systems. This produces a framework for understanding differential viability rather than an empirically tested classification. We will revise to include brief illustrative scenarios for each category and articulate potential falsifiable thresholds (e.g., regulatory intensity scores above which Buy remains dominant) to make the mapping more explicit and support future empirical work, without changing the core conclusions. revision: yes

Circularity Check

0 steps flagged

No circularity: conceptual synthesis of external theories and qualitative assessment

full rationale

The paper performs a factor-level analysis by applying pre-existing transaction cost economics (Williamson) and resource-based view frameworks to a qualitative assessment of current agentic AI capabilities. The seven decision determinants are re-evaluated interpretively without any fitting of parameters, self-definitional loops, or self-citations that bear the central load. The typology of enterprise applications and the conclusion that the Make option shifts to a hybrid governance form are derived from these external foundations rather than reducing to the paper's own inputs by construction. No equations, fitted predictions, or ansatzes are present; the derivation chain is self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

The claims depend on two background assumptions: that transaction cost economics and resource-based view remain valid lenses for AI-augmented decisions, and that current AI capabilities can be assessed qualitatively to predict directional shifts in the listed determinants. No free parameters or invented entities are introduced.

axioms (2)

domain assumption Transaction cost economics and the resource-based view of the firm apply directly to decisions involving agentic AI for software development.
Invoked as the conceptual foundation for re-evaluating the seven determinants.
ad hoc to paper Current agentic AI capabilities can be assessed qualitatively to determine how they reshape cost, specificity, lock-in, and other factors.
The paper states it combines an assessment of current AI capabilities with the economic frameworks.

pith-pipeline@v0.9.0 · 5552 in / 1453 out tokens · 68250 ms · 2026-05-07T10:45:47.812155+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

2 extracted references · 2 canonical work pages · 1 internal anchor

[1]

illogical

Accessed March 6, 2026. Kif Leswing. AI fears pummel software stocks: Is it “illogical” panic or a SaaS apocalypse? CNBC, February 2026. Published February 6, 2026. Bloomberg News. What’s behind the “SaaSpocalypse” plunge in software stocks. Bloomberg, February 2026. Published February 4, 2026. Anthropic. Claude Code: Agentic coding tool. Anthropic, Febru...

work page doi:10.1086/227496 2026
[2]

Agentic Much? Adoption of Coding Agents on GitHub

Open-source LLM orchestration framework. Launched 2022; includes LangGraph for agent workflows. https://github.com/langchain-ai/langchain. Romain Robbes, Théo Matricon, Thomas Degueule, Andre Hora, and Stefano Zacchiroli. Agentic much? adoption of coding agents on GitHub.arXiv preprint arXiv:2601.18341, 2026. doi:10.48550/arXiv.2601.18341. Musfiqur Rahman...

work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2601.18341 2022