AI is moving from experimentation to practical use. Location is what makes that use meaningful in the real world.
That was one of the clearest messages from the Esri User Conference (UC) 2026. Across the plenary, AI in ArcGIS sessions and customer demonstrations, the conversation moved beyond whether organisations should use artificial intelligence. The more important question was how to use AI in ways that are grounded in trusted data, spatial context and real operational workflows.
For Singaporean organisations managing infrastructure, networks, services, communities and risk, that shift matters. AI can generate content, analyse information and automate tasks, but it does not automatically understand where something is happening, what is nearby, how assets and communities are connected, or what action is appropriate. GIS provides that missing context.
Here are five takeaways from Esri UC 2026 for organisations exploring AI in GIS.
1. AI needs location to understand the real world
Most enterprise decisions have a location dimension. A service interruption affects a particular network and community. A new development changes demand on nearby roads and utilities. A hazard becomes operationally significant because of the assets, people and access routes in its path.
AI may identify a signal, summarise a report or recommend a next step. GIS adds the relationships that make those outputs useful: location, time, proximity, terrain, networks, assets and events. It connects a broad answer to the specific conditions on the ground.
The value runs both ways. AI helps GIS teams extract information at scale, detect patterns, streamline workflows and make complex tools easier to use. GIS helps AI produce answers that are geographically aware and better aligned to real-world decisions.
2. ‘AI in GIS’ is not one capability
Esri UC framed the AI in ArcGIS roadmap around three complementary areas. Understanding the difference helps organisations choose the right approach for the problem they need to solve.
- AI tools and models apply machine learning, deep learning and other analytical methods to geospatial data. These capabilities can classify imagery, detect objects and change, make predictions and reveal spatial patterns at a scale that would be difficult to achieve manually.
- AI assistants use conversational experiences and generative AI to help people work with ArcGIS. They can support tasks such as creating expressions, generating code, documenting content or navigating workflows, reducing friction while keeping the user in control.
- Agentic AI and integration connect GIS data, tools and services with agents and wider enterprise AI ecosystems. The aim is to support multi-step, context-aware workflows that can interpret a request, select appropriate tools and take governed action.
Esri's Trusted AI in ArcGIS framework uses the same three-part view. It is a useful reminder that an AI strategy for GIS should not begin with a single model or assistant. It should begin with the problem, the workflow and the level of human oversight required.
3. GeoAI is expanding from task-specific models to foundation models
GeoAI has long helped practitioners apply AI techniques to spatial data, such as detecting assets in imagery, classifying land cover or forecasting spatial outcomes. At Esri UC 2026, the notable shift was the growing role of geospatial foundation models: models trained at scale that can be adapted across multiple related tasks.
New capabilities announced around UC included location encoders that create numerical representations, or embeddings, of places. Esri's Global Location Encoder learns from Sentinel-2 imagery, while the USA Geodemographic Foundation Model draws on demographic, socioeconomic, housing and environmental information. These embeddings can support similarity searches, clustering, predictive modelling, interpolation and site selection.
Esri also introduced GeoVLM, a geospatial vision-language model designed for Earth observation. It connects remote-sensing imagery with natural-language prompts and is intended to support tasks including object detection, segmentation, classification, image captioning and visual question answering. GeoVLM was announced for private beta, so it should be viewed as an emerging capability rather than a production-ready default.
The opportunity is significant, but foundation models still need careful evaluation. Teams need to understand the source data, geographic coverage, intended use, accuracy, limitations and explainability of a model before applying its output to important decisions.
4. Assistants will reduce friction, but expertise remains essential
Some of the most practical demonstrations at UC involved AI assistants embedded within familiar GIS experiences. Instead of asking users to move data into a general-purpose chatbot, these assistants work with the language, tools and context of ArcGIS.
Examples included generating Arcade expressions, SQL queries and Python code from natural-language instructions; helping users create item descriptions and metadata; and turning a plain-language analytical request into code that a user can review before running.
The value is not only speed. Assistants can reduce the effort involved in routine tasks, help experienced practitioners explore unfamiliar capabilities and make spatial information more accessible to people outside the GIS team.
But accessibility is not the same as autonomy. The strongest operating model keeps human judgement close to the work. GIS professionals remain responsible for checking assumptions, validating outputs, understanding spatial bias and deciding whether a result is fit for purpose.
5. Agentic AI could bring GIS into more enterprise workflows
The longer-term shift is from assistants that support individual tasks to agents that can coordinate multiple steps across systems. For GIS, this could mean an agent recognising that a question has a spatial component, finding the right location services, querying authoritative information and bringing the result into a broader business workflow.
Model Context Protocol (MCP) is one pathway for enabling this connection. Esri released MCP support for ArcGIS Location Services in beta in June 2026, initially exposing capabilities including geocoding, routing, elevation and static maps to compatible AI agents.
The business implication is bigger than a new interface. It creates a path for geographic knowledge to be used inside enterprise AI systems, rather than remaining isolated in specialist applications. A logistics agent could consider routes and travel times. A service agent could identify the assets and communities affected by an incident. A planning workflow could combine organisational data with authoritative location context before recommending action.
This is also where governance becomes critical. Agents that can act require clear permissions, authoritative inputs, auditability, security controls and defined points of human approval.
From a system of record to a trusted system of action
The direction outlined at Esri UC 2026 was not AI for its own sake. It was about the evolution of GIS: from documenting what and where, to helping organisations understand what is changing and coordinate what happens next.
One demonstration showed how the United States Census Bureau is using AI-powered imagery change detection to identify new housing and development, maintain geographic datasets and focus human effort on more complex decisions. It is a useful example of where AI in GIS creates value: a defined operational problem, authoritative data, repeatable analysis and people applying judgement to the result.
For organisations considering their next step, the lesson is clear: start with a workflow, not a technology category. Identify a high-value decision where location matters, establish the data and governance required, and determine where AI can remove friction or increase scale without weakening accountability.
What should Singaporean organisations do next?
- Prioritise a real operational use case. Focus on a recurring decision involving assets, networks, services, risk or communities.
- Assess data readiness. AI cannot compensate for incomplete, outdated or poorly governed spatial data.
- Choose the right AI pattern. Decide whether the need is analytical GeoAI, an in-product assistant, or integration with a broader agentic workflow.
- Define human oversight. Make it clear who reviews outputs, approves actions and remains accountable for the outcome.
- Measure operational value. Track time saved, improved accuracy, faster response, reduced risk or better service – not simply the number of AI features adopted.
The organisations that gain the most from AI in GIS will not be those that deploy the most AI. They will be those that use trusted technology, geographic context and human expertise to solve real problems, then scale what works.
Discover the major ArcGIS announcements, demonstrations, and strategic insights from Esri UC 2026 in our regional wrap-up webinar.