AI assistants are natural-language interfaces inside ArcGIS applications that help users complete tasks or write code. They are built on LLMs and geospatial agents, grounded on ArcGIS content, and scoped to the user’s permissions. This chapter covers the two types of assistants (task and coding), how they work with the user’s data, and related architectural considerations.
Task assistants help users complete GIS tasks more easily and quickly. They operate within ArcGIS applications, guiding users through actions, answering questions, and triggering operations on the user’s behalf.
The ArcGIS Pro assistant is the most feature-rich task assistant. It can explore ArcGIS Pro documentation, generate Query Layer SQL expressions, generate openCypher queries for knowledge graphs, and trigger common ArcGIS Pro actions through ArcPy and Arcade code generation. It is broad in scope, covering data management, analysis, mapping, and administration tasks within Pro. It is also extensible through the ArcGIS Pro SDK (.NET), allowing organizations to register custom assistant extensions that add domain-specific capabilities.
Other task assistants are integrated across ArcGIS applications. The Business Analyst assistant guides users through site selection, market analysis, and demographic workflows. The Survey123 assistant helps survey authors design and configure forms. The Item Details Assistant helps content authors describe and document items, improving both discoverability and AI-readiness for downstream workflows.
Additional task assistants are available or in development for ArcGIS StoryMaps, Hub, Instant Apps, Field Maps, Drone2Map, Solutions, ArcGIS for Microsoft Teams, and ArcGIS documentation. The inventory continues to expand as Esri integrates assistant capabilities across the ArcGIS platform.
Coding assistants help users generate, understand, and troubleshoot code. They lower the barrier to scripting and automation for users who are less comfortable with code, and they help experienced developers move faster on repetitive tasks.
The ArcGIS Notebooks assistant generates, fixes, and explains Python code within the ArcGIS Notebooks environment. It operates in two tiers. Standard mode handles administrative tasks and light data engineering, while Advanced mode supports more complex analytics and large-scale data workflows. Both modes are grounded on ArcGIS Python APIs, documentation, and coding patterns.
The Arcade assistant helps users write Arcade expressions for pop-ups, visualizations, labeling, attribute rules, and other contexts where Arcade is used. Since Arcade is embedded across many ArcGIS applications, this assistant has broad applicability for configuration workflows that do not require full Python scripting.
AI assistants in ArcGIS are grounded on the content and context relevant to their scope. This grounding is what differentiates them from a generic chatbot; they respond using information from ArcGIS, not just their training data.
Each assistant is grounded differently depending on its context. The ArcGIS Pro Assistant draws on documentation, the application state, and the user’s project data. App-specific task assistants draw on the content from within their application scope. Coding assistants draw on API documentation, coding patterns, and the current notebook or code context.
Perhaps most importantly, assistants work within the data and tools the user has access to. Existing identity and access controls in ArcGIS extend naturally to assistant interactions. An assistant cannot see content the user cannot see and cannot invoke operations the user is not authorized to perform. This means the existing security model governs what assistants can do without requiring a separate permission layer.
Data used with AI assistants remains private to the organization. Prompts and data are not used to train AI models without explicit authorization, and customers retain ownership of their prompts and data. When third-party AI services are involved (as they are today through the LLM gateway), Esri uses enterprise-class AI instances to segment and protect customer data from other customers. These principles are documented in detail in the ArcGIS Trust Center.
The effectiveness of assistants depends directly on the quality of the data they can access. Well-described items with complete metadata, clear field descriptions, and meaningful tags produce better assistant responses than poorly documented content. The AI-readiness practices described in geospatial data for AI apply directly here, and investing in data quality and documentation pays dividends when working with assistants.
AI assistants introduce several architectural considerations beyond those of traditional ArcGIS applications.
Progressive rollout through role privilege. A dedicated role privilege for AI assistants lets administrators control which users have access. This enables organizations to take a staged approach: enable assistants for specific roles or groups, observe usage and behavior, then expand access as confidence grows. Assistants are not enabled by default; administrators or users must opt in.
User type licensing. AI assistant capabilities vary by user type. Higher user type levels include more assistant capabilities. Architects designing solutions should account for which assistants and features are available at each licensing level, particularly when planning organization-wide rollouts.
LLM hosting and connectivity. AI assistants today depend on LLMs hosted through ArcGIS Online, which serves as the LLM gateway. This means assistants require connectivity to ArcGIS Online to function, and are not currently available in fully disconnected or air-gapped environments. Self-hosted LLM support is on the roadmap for ArcGIS Enterprise, which will provide a path for customers who need to operate assistants in sovereign or disconnected environments. The scope of self-hosted LLM support is still being defined. See integration and interoperability for more detail on the LLM gateway model and deployment considerations.
Cost and consumption. Some assistant interactions may consume credits in ArcGIS Online. Monitoring assistant usage is part of the broader observability story covered in architecture pillars. Learn more about credits in ArcGIS Online.
Extensibility. The ArcGIS Pro SDK allows developers to write and register custom assistant extensions in .NET, adding domain-specific capabilities to the ArcGIS Pro assistant. This means organizations can extend the assistant’s scope beyond what Esri provides out of the box, for example adding custom tool invocations, domain-specific knowledge, or organization-specific workflows. Custom extensions inherit the same security model as the core assistant.