The rise of large language models (LLMs) and large multimodal models (LMMs) has enabled agentic systems that combine natural language understanding with tool-based execution. In geographic information systems (GIS), this shift is turning traditional expert‑driven workflows into semi‑autonomous pipelines that can interpret user intent, construct spatial workflows, and run geospatial analyses. Existing approaches, however, suffer from fragmented integration of reasoning, execution, and evaluation, especially in complex real‑world settings.
This paper surveys recent agentic GIS frameworks, benchmarks, and surveys, highlighting limitations in spatial reasoning, execution robustness, validation, governance, and evaluation. Building on these insights, we introduce ANASSA (Autonomous Neural Agents for Spatial Systems Architecture), a unified agentic AI orchestration framework that incorporates structured spatial reasoning, multi‑agent workflow orchestration, execution feedback, authoritative spatial validation, provenance, uncertainty handling, and human decision authority.
ANASSA’s architecture specification consists of four layers, eleven components, and a six‑step Geospatial AI Cognitive Loop. Cross‑component contracts define interactions, while governance mechanisms ensure traceability, reproducibility, and accountability. The layers span perception, reasoning, execution, and governance; key components include intent parsing, spatial semantic graphs, task scheduling, execution monitoring, and result validation.
The paper does not present empirical performance results; instead, it provides a comprehensive design blueprint for future implementation and deployment studies, aiming to advance agentic GIS systems toward greater reliability and transparency.
Review