From tool to collaborator: rethinking GIScience as GIS becomes agentic and autonomous

Geographic Information Systems (GIS) have traditionally been positioned along Wright, Goodchild and Proctor’s continuum of GIS as tool, toolmaking, and GIScience, a framework that treats GIS as primarily executing human-authored intent. This paper argues that agentic and autonomous GIS strains that assumption by moving GIS beyond predefined command execution toward intent interpretation, workflow construction, and iterative procedural revision. Through an integrative narrative review, we trace a shift in intent from expert command execution to partial delegation of planning and procedural assembly to systems operating in iterative human-agent workflows. We propose the Agentic Role Entanglement (ARE) model to conceptualize this shift. ARE describes an epistemic action loop in which tool use, runtime method construction, and epistemic inquiry may become coupled within a single analytic episode. The model yields falsifiable implications for evaluation and peer review, motivating episode-level validation when methods are generated or revised at runtime. We then outline a validation stack centered on provenance, reproducibility, workflow, code verification, geospatial sanity checks, and uncertainty-bounded claims, and connect these requirements to a lifecycle view of responsibility and ethical governance. We conclude by advocating a post-instrumental GIScience that treats human-agent systems as the unit of analysis for scientific defensibility, transparency, and accountability in autonomous GIS workflows.