MARE: An Anti-Obsolescence Meta-Architecture for Sustainable and Sovereign AI Systems Abstract Artificial intelligence evolves faster than many organizations can afford to follow. Models, libraries, APIs, runtimes, cloud services, and hardware platforms are continually deprecated, discontinued, or constrained by upgrade cycles beyond the control of any individual system. As a result, a component may remain fit for purpose yet become unusable because the surrounding technological ecosystem has changed. For well-resourced institutions, this is a recurring cost; for non-profits, schools, libraries, municipalities, and other public-interest organizations, it can become a barrier to technological participation. This paper argues that this problem is fundamentally architectural and introduces MARE (Meta-Architecture for Role-Executor AI Systems), a technology-agnostic model for preserving functional continuity under technological change. MARE defines systems through stable Roles, architectural responsibilities that specify what the system must do and are characterized by goals, contracts, parameters, constraints, and policies. Recursive Role decomposition captures internal organization without binding the system to a specific implementation stack. Executors are replaceable and upgradable enactment units that realize Role responsibilities at a given time through models, services, legacy modules, human processes, or nested Roles. Interactions among Roles are organized through Transfer, Activation, and Observation (TAO) semantics, respectively supporting information exchange, Role activation, and monitoring. The TAO triad enables Roles to exchange data and coordinate behaviour through heterogeneous architectures, technologies, and integration interfaces. By delegating realization to Executors, MARE localizes technological aging rather than propagating it through the architecture. Components from different generations, cost profiles, and maintenance conditions can therefore coexist within the same system, supporting conditions for architectural sustainability, technological sovereignty, and accessibility under diverse resource constraints. MARE emerged from MGVT, an implemented multimodal generative AI tutoring system, through a bottom-up process of architectural abstraction. Rather than treating MGVT as a domain-specific tutoring application, MARE distills from it a generalized model for AI systems exposed to ecosystem volatility, in which Role stability, Executor replaceability, TAO interaction semantics, and recursive encapsulation are co-defined as architectural mechanisms for maintainability, heterogeneity, and temporal interoperability. Keywords: software architecture; meta-architecture; AI systems; software evolution; technological obsolescence; architectural anti-obsolescence; Role-Executor architecture; Executor replaceability; architectural event semantics; TAO semantics; technological sovereignty; architectural sustainability; sustainable computing; legacy integration; temporal interoperability.
Giordano et al. (Sun,) studied this question.