Design practice frequently characterises generative artificial intelligence (AI) as a compliant, tool-like extension of human intention; however, recent reports of atypical behaviours motivate a reexamination of this metaphor. Grounded in Object-Oriented Ontology, agential realism, and posthumanist theory, this paper introduces the Tensional-Field Relational Framework (TFRF) for human–AI co-creation. TFRF models collaboration as a flat network comprising five semi-opaque node classes: human designer, AI system, creative artefact, intentional vector, and sociotechnical environment, all linked by dynamically weighted edges and circulating flows. We operationalise this perspective as a diagnostic governance grammar: relations are instrumented as edges with quantified flows (data movement, permission transitions, reward cadence, temporal dynamics) and governed at Obligatory Passage Points (OPPs). A trace-based procedure reconstructs translation chains (prompts → platform mediation → model settings → versioned artefacts), thereby rendering “orphaned edges” and “uncontrolled flows” auditable and locating concrete controls (provenance, consent, and variability) at specific OPPs. The contribution is a reproducible lens that advances beyond conceptual assembly toward actionable diagnostics for risk identification and responsibility allocation in contemporary cocreation.