Agentic AI introduces orchestration, automation, and integration through agents and agentic workflows. This is the newest of the three geospatial AI capabilities and the fastest moving. It is also the most architecturally significant, because it changes how ArcGIS connects to the broader enterprise AI ecosystem.
Agentic AI in ArcGIS follows two complementary architectural patterns, and most organizations will eventually use both.
Geo-centric workflows bring agentic AI into GIS. This includes agentic mapping applications and geospatial agents built within ArcGIS. The GIS environment is the center of gravity, and the agents serve its workflows. A utility company deploying a geospatial agent to automate network tracing and validation is working in a geo-centric pattern.
Geo-enabled workflows bring GIS into broader agentic ecosystems. Here an external agent platform orchestrates the work, and ArcGIS participates as a data, tool, and/or agent provider. The center of gravity is outside of the GIS, often a cloud or enterprise-level agentic platform, and ArcGIS contributes geospatial capabilities that the external system lacks natively. A logistics company whose supply-chain agent calls ArcGIS for routing and geocoding through MCP is working in a geo-enabled pattern.
These two patterns are not in competition. They both serve to make geospatial intelligence available wherever decisions are being made. The architectural axis is how agents are orchestrated: in a geo-centric pattern, ArcGIS concepts and users lead the workflow definition, whereas in a geo-enabled system, the users and capabilities of that other system drive the decisions. Both patterns need identity propagation, access control, observability, and governance, but the specifics differ depending on which side of the boundary the design sits. The integration and interoperability chapter covers the architectural details.
Most AI systems and agent platforms have no native geospatial capabilities. They can call APIs, read databases, and generate text, but they cannot geocode an address, calculate a service area, understand topological relationships, or produce a map without external help. Additionally, many organizations have rich, authoritative geospatial data in their GIS they want to expose to agents. ArcGIS fills this gap by providing geospatial intelligence to external agents.
One way that ArcGIS exposes its capabilities to external agents is through MCP (Model Context Protocol). MCP is an open interoperability standard that defines how AI models and agents connect to external data sources and tools. By supporting MCP natively, ArcGIS makes it possible for any MCP-compliant agent to call geospatial operations, such as geocoding, routing, spatial queries, geoprocessing, content search, map export, geoenrichment, elevation, places, and more, without the agent platform needing to understand GIS internally. Beyond MCP services, the REST APIs and methods exposed by ArcGIS can also be used by other agentic systems, either as a direct user prompt or as part of a well-defined skill.
MCP support is rolling out across ArcGIS Enterprise, ArcGIS Online, and ArcGIS Location Platform. Each deployment target has its own rollout cadence and initially exposes a different set of capabilities. The set of supported operations will expand over time. For organizations that need additional or domain-specific agentic integration for their specific needs, Esri Professional Services and partners can deliver integration through APIs and custom solutions.
The architectural depth of MCP integration, including what is exposed, how authentication works, how to govern agent traffic, is covered in integration and interoperability.
Agentic mapping applications use AI agents to expand the experience of mapping through natural-language interaction. Rather than requiring users to know which tools to run or how to configure a query, an agentic mapping application lets them describe what they need in plain language and the agent figures out how to deliver it.
Esri provides two paths for building these applications.
Configurable templates offer a low-code path. The Data Explorer template for ArcGIS Instant Apps lets organizations configure a web map with an agentic chat interface. Users can ask questions about the data, filter and query features, and explore spatial patterns through conversation rather than through traditional GIS tool interfaces. The Data Explorer template is in beta.
Developer components offer a full-code path. AI components in the ArcGIS Maps SDK for JavaScript give developers the building blocks for custom agentic mapping applications: chat interfaces, out-of-the-box agents for navigation and data exploration, and the ability to build custom agents with domain-specific logic. These components are in beta. To learn more, see an Introduction to building agentic mapping applications with ArcGIS Maps SDK for JavaScript.
Both paths follow the same architectural pattern, they are web applications that embed an agent and ground it on the organization’s spatial content. They inherit the same security, identity, and data access model as other ArcGIS web applications.
One note on building practices. AI-assisted code generation can accelerate prototyping of agentic mapping applications significantly. And many of the ArcGIS Maps SDKs can be used for AI-assisted software development. Keep in mind that production applications require the same engineering discipline as any production system, including code review, testing, security audit, performance validation, and observability. AI code generation can be a development accelerator, but is not a substitute for production architecture and engineering.
Geospatial agents are agents designed specifically for geospatial tasks. They use the same components as any agent (LLM, instructions, tools, knowledge, memory), but their tools are GIS tools, their knowledge is geospatial content, and their memory holds spatial context. They incorporate geospatial reasoning and skills to accomplish multi-step work and advanced workflows that previously required manual orchestration.
Esri is building geospatial agents through two paths.
Prebuilt agents are provided by Esri and are ready-to-use for common geospatial workflows, such as navigation assistance, data exploration, and spatial Q&A. Prebuilt agents are currently in beta and exposed through ArcGIS Maps SDK for JavaScript. They give organizations a starting point for agentic experiences without requiring custom agent development.
Custom agents through Agent Builder. ArcGIS Agent Builder is a forthcoming visual tool for designing geospatial agents without coding. It will allow organizations to configure custom agents by specifying instructions, connecting tools, pointing to knowledge sources, and defining boundaries — producing agents that can be deployed in ArcGIS applications or exposed to external systems. Agent Builder is currently in an alpha release stage and broader access is planned for the near future.
Both paths are early in their definition and best practice identification. Organizations planning for geospatial agents should understand the direction and begin thinking about use cases, data readiness, and governance. It is recommended to plan deployments around the capabilities intended for beta or GA, not those that are still in alpha.