Geospatial AI is the integration of GIS and AI and represents all AI capabilities Esri builds into ArcGIS.
Geospatial AI in ArcGIS is organized around three capability areas. Each is covered in its own chapter later in the lens; this chapter describes their value proposition and alignment to the architecture of ArcGIS.
AI tools and models automate data creation, extraction, and analysis. They are powered by machine learning, deep learning, and advanced spatial statistics, and they let users extract information from imagery, 3D data, and text, detect patterns, and make predictions from complex datasets. These are the workhorses of applied geospatial science, delivering capabilities that help GIS analysts and data scientists accelerate analytical work at greater scale. AI tools and models are the most mature of the three capabilities, with over a decade of investment behind them.
AI assistants are in-app natural-language interfaces that make GIS more accessible and its users more productive. Task assistants guide users through workflows in apps like ArcGIS Pro, Business Analyst, and Survey123. Coding assistants help users generate and troubleshoot Python and Arcade. Together they lower the barrier for new users while helping experienced practitioners move faster. They are backed by geospatial agents and grounded on the organization’s own ArcGIS content.
Agentic AI introduces orchestration, automation, and integration through agents and agentic workflows. Agents can perceive context, reason about goals, and act autonomously or with a human in the loop. In ArcGIS, agentic AI shows up in three places: through integration with third-party agent platforms (where ArcGIS acts as a tool provider via MCP or other mechanisms), through agentic mapping applications (where natural-language interaction drives spatial insights), and through geospatial agents built within ArcGIS itself. This is the newest and fastest-moving of the three capabilities.
The three are not isolated. A single solution might use AI tools and models to extract features from imagery, surface those results through an AI assistant for review, and hand off validated outputs to an agent for downstream distribution.
Geospatial AI delivers value to organizations beyond what GIS or AI can offer on their own.
The geospatial gap in enterprise AI. Most enterprise AI systems are not trained to work with geographic science or extract location-based context. They can reason about text, tabular data, images, and videos; however, they typically cannot answer questions that depend on where things are, how they relate spatially, or how spatial patterns influence outcomes. Geospatial AI bridges this gap by connecting trusted GIS technology with the reasoning power of modern AI.
Automate geospatial workflows. Routine tasks that previously required significant time from specialized staff, such as feature extraction, classification, image inspection, and data QA, can be automated or substantially accelerated. This frees teams to focus on interpretation, validation, and higher-value analysis rather than repetitive production work.
Accelerate analysis at scale. Geospatial AI makes it practical to analyze volumes of imagery, text, and sensor data that would be infeasible by traditional means. Regional-scale land cover classification, national infrastructure monitoring, and pattern detection across millions of records become tractable problems rather than multi-year projects.
Expand access to spatial insights. Assistants and agents make it possible for people without deep GIS expertise to ask spatial questions, find the right data, choose appropriate workflows, and act on results. They also help users learn GIS workflows faster, which broadens the reach and impact of an organization’s geospatial investment beyond the GIS team.
Geospatial AI does not replace the existing ArcGIS architecture. It extends it.
The base ArcGIS architecture is organized in three tiers: applications, services, and data. Each tier includes both Esri-built and organizational (user/enterprise) built components, for example, Esri delivers broad data and content via Living Atlas whereas an organization brings their own data and content to ArcGIS to use in combination with Esri’s data. AI extends this model by adding a new horizontal layer for agents (including Esri-built agents, and customer-built agents) and by introducing AI capabilities into the existing tiers.
Where the three capabilities show up in the architecture:
AI tools and models span all three tiers. Pre-trained and foundational models align with data tier, with Esri-built components provided through Living Atlas. Tools and capabilities align with services tier, including geoprocessing services, AI-powered services, and inference endpoints. Experiences for working with AI tools and models are provided through the application tier, including ArcGIS Pro, ArcGIS Notebooks, and ArcGIS Deep Learning Studio.
AI assistants live in the application tier, surfacing throughout the user experience of ArcGIS applications as in-app interfaces. AI assistants depend on services, models, and data to function.
Agentic AI introduces the new agents layer and opens ArcGIS to external agentic systems through standards-based interoperability (MCP, A2A). Agents orchestrate tools in the services tier, consume data from the services and data tiers, and surface results through applications, including Esri provided applications as well as user-built applications.
ArcGIS plays two roles in the AI ecosystem. As an AI consumer, it uses AI internally: LLMs power assistants and agents, ML and DL models power tools. As an AI producer, it exposes AI capabilities — tools, services, geospatial content, and agents — for use by external AI systems. An organization’s broader AI strategy may rely on ArcGIS in either role or both, and the architectural decisions differ depending on which side of that boundary the design sits.
Finally, geospatial AI introduces several architectural implications that shape every design decision in the chapters ahead:
Non-determinism. AI workloads, particularly those involving LLMs and agents, can produce different outputs from the same input. Systems designed around them need to handle variability rather than assuming repeatable results.
New integration surfaces. MCP and A2A create new boundaries where ArcGIS connects to the broader AI ecosystem. These surfaces need careful review and implementation of authentication, authorization, observability, and governance.
Data quality expectations. AI amplifies the consequences of poor data. Grounding, training, and inference all depend on data that is accurate, well-documented, and governed. The bar for data quality in AI-enhanced systems is higher than it has been for many traditional GIS workflows — a topic explored in depth in the Geospatial data for AI chapter.