Work Without Objects: The Epistemic Capital of Software Engineering

Abstract

Software engineering has converged on a set of near-universal practices — version control, continuous integration, code review, on-call rotation, staged deployment, open-source dependency — none of which appears in the field's foundational theory or in the theories of work it inherited. This paper argues the convergence is not immaturity but selection under a substrate that supplies no natural coarse-graining. Physical engineering rests on the ability to replace microscopic models with macroscopic ones that compose. The physical substrate supplies this for free. The computational substrate, because it is universal, supplies none: no natural distance, no dissipation, no stable kinds. Every engineering structure — every boundary, module, type, component — is therefore a constructed theory, maintained by people, adjudicated only by execution. Two consequences follow. First, the industrial ontology (units, batches, tolerances, inspection, the plan) has no substrate-supplied referent in software and cannot be made to fit by further professionalisation. Second, every engineering decision on this substrate moves two balances simultaneously: functional capital (new capabilities) and epistemic capital (evidence-grounded confidence in those capabilities). The industrial frame tracks only the first, making the second invisible — and systematically recommending decisions that spend it. Epistemic capital is not a quantity but a region: the evidenced envelope, indexed to a system and the operational context under which evidence was obtained. It grows through non-events (unfalsified executions) and is destroyed by single witnesses. Its infrastructure — CI pipelines, test suites, telemetry, monitoring — appears on an invoice as overhead while constituting half the product. A fielded software system is a sociotechnical system: code, evidential infrastructure, and the people who hold the theory of how these relate. Remove any element and what remains is not a degraded product but an uncharacterised one. The paper proposes a counter-ontology whose primitives are relationships between observers and executions rather than properties of artefacts, and uses it to re-examine practices the field takes for granted: why test suites never shrink, why "not invented here" can be rational, why technical debt is capital depletion rather than deferred work, why Conway's law is an architectural constraint rather than a sociological curiosity, and why tiger teams are a structural response to coupling the hierarchy was not designed to see. A witness-cost quadrant distinguishes the two coherent engineering strategies (substrate restriction and evidential investment) from the symbolic interventions that address neither the dynamics nor the evidence. The paper draws on formal results established in eight companion papers (CHP, CCI, SSEP, TWS, BPS, CSG, TAL, AST) by the author but is intended to be readable independently of them. Version history v1.2 (September 2026) – Substantial revision. Expanded the SDV case study with a three-phase historical structure (disconnection → digital connectivity → shared substrate). Added §5.5 on the evidenced envelope as the actual form of epistemic capital, §5.6 on the build-vs-import decision, and infrastructure-cost and security perspectives. Restructured §5 into subsections. Replaced the refinement-order argument with concrete examples and corrected the code/cognitive-shadow framing. Grounded the organisational-form discussion in Lawrence & Lorsch and Galbraith. Added a product-definition claim to the Conclusion. Replaced "observation map" with "abstraction" throughout. Simplified physics language, glossed domain abbreviations, and reduced meta-commentary for accessibility. Added Boeing Starliner CFT investigation as a cited instance. Removed the underdeveloped "Brooks, corrected" subsection. 46 references. v1.0 (August 2026) – Initial preprint. Keywords: software engineering, epistemic capital, coarse-graining, computational substrate, bounded local theory, software-defined vehicle, organisational design, Conway's law, continuous integration, evidence-based engineering, software safety, non-industrial technology, gauge hypothesis, witness, leakage