Abstract
Advanced AI can increasingly inspect software interfaces, infer intent, translate between heterogeneous systems, and generate integration logic on demand — raising a provocative question: do standards still have a function when machines can mediate interoperability dynamically? Decomposing standards into their two separable jobs — compressing the cost of compatibility, and fixing the locus of commitment — this paper argues that AI collapses the first and cannot, by construction, supply the second, because commitment is a relation between accountable persons, not a property of representations. Using tokenized finance as the limiting case and analysing sixteen operational domains, it projects the future forms of standards and concludes that AI redistributes the standards function rather than retiring it: expect fewer, deeper, more executable standards. Translation becomes cheap; commitment remains scarce.
Executive summary
Advanced AI is approaching the capacity to inspect software interfaces, infer intent, translate between heterogeneous systems, detect ambiguity, and generate integration logic on demand. That raises a provocative question: do standards retain a function when machines can mediate interoperability dynamically? The paper steelmans the provocation before answering it.
The question dissolves once standards are decomposed into their two separable functions:
- They compress the cost of compatibility — and advanced AI collapses this function.
- They fix the locus of commitment — and AI cannot, by construction, supply this, because commitment is a relation between accountable persons, not a property of representations.
Using tokenized finance as the limiting case, the paper argues that the binding constraint on institutional tokenization has never been syntactic interoperability but the absence of shared institutional meaning around rights, controls, obligations, lifecycle events, finality, and accountability. It analyses sixteen operational domains — from data mapping to model governance — distinguishing in each what AI can plausibly automate from what still requires standards, governance, legal authority, or market convention.
Finally, it projects the future forms of standards: shared ontologies with formal semantics, executable specifications, conformance test suites, control-property standards, machine-readable law, agent mandate and negotiation protocols, evidence-grade audit logs, settlement and finality standards, and certification regimes for models and adapters.
The conclusion is a redistribution of the standards function, not its retirement: AI changes the economics of making and enforcing standards far more than it changes the need for having them. Expect fewer, deeper, more executable standards. Translation becomes cheap; commitment remains scarce.