Contemporary crises, ranging from pandemics and climate-related disasters to geopolitical conflicts, are increasingly interconnected and transboundary, giving rise to polycrises characterised by cascading effects, high uncertainty, and fragmented information. These dynamics challenge traditional crisis management and call for coordination mechanisms capable of aligning heterogeneous stakeholders under time pressure and asymmetric knowledge. Although Artificial Intelligence (AI), particularly generative AI, is gaining prominence in crisis management, it is still predominantly framed as a tool for prediction or automation. This paper argues that such a view overlooks AI’s broader potential as a socio-technical coordination mechanism that augments human capabilities by supporting shared sensemaking and knowledge integration across organisational and institutional boundaries. Adopting a design science research approach combined with conceptual development, this paper proposes a reference framework that conceptualises AI as an embedded coordination infrastructure. The framework integrates three interrelated components. First, it specifies five AI-enabled coordination roles (Sentinel, Broker, Verifier, Orchestrator, and Scribe) aligned with core knowledge management processes (sensing, integration, validation, application, retention). These roles are derived by linking coordination dependencies (flow, fit, share) to knowledge processes that become critical during polycrises and are conceived as overlapping mechanisms embedded in human-AI interaction and organisational practices. Second, it proposes a four-layer architecture that operationalises these roles across heterogeneous environments, spanning data/provenance, knowledge processing, coordination enablement, and governance and learning. Third, the study introduces the construct of AI-enhanced regenerative coordination capacity, articulated through three pillars of crisis readiness: dependency visibility, coordination mechanism agility, and collective learning and institutionalisation. To make the concepts concrete without claiming empirical validation, the framework is examined through structured vignettes derived from recurring crisis patterns and assessed via an analytical, vignette-based evaluation. A common specification template supports traceability from roles to layers to outputs and enables assessment of coverage, governability (provenance, contestability, bounded authority), and explanatory usefulness. Practically, the framework informs the design of interoperable, transparent, and adaptive crisis coordination systems that strengthen informed trust, organisational learning, and cross-boundary collaboration in polycrisis settings.
