[vc_row][vc_column width=”1/1″][vc_column_text]Esri is bringing AI to GIS. Eagle is bringing GIS to AI.[/vc_column_text][us_separator size=”small”][vc_row_inner][vc_column_inner width=”1/1″][vc_column_text]
[/vc_column_text][/vc_column_inner][/vc_row_inner][us_separator][vc_row_inner][vc_column_inner width=”1/1″][vc_column_text]Esri’s roadmap is steadily embedding AI assistants into ArcGIS, Arcade Assistant, ArcGIS Pro Assistant, Survey123 Assistant, and more, arriving through 2026 and beyond. It’s a genuine step forward for anyone working inside ArcGIS.Esri has been steadily been bringing AI capabilities to ArcGIS across all aspects of the platform, from the Geospatial tools and models that let us analyse imagery and extract objects, to AI powered assistants to help us write arcade code through assistants in applications like Survey123.[/vc_column_text][us_separator size=”small”][us_image image=”4499″ align=”center” link=”%7B%22url%22%3A%22%22%7D”][vc_column_text]
Graphic depicting potential usage of Geospatial AI
[/vc_column_text][us_separator size=”small”][vc_column_text]Eagle has been focusing on how we can bring GIS capabilities to AI. How can we move beyond simple chatbots over our geodatabases to create tools and agents that actually make users more productive, provide location analytics to the non specialist and provide the analysts tools help them do their job more efficiently and effectively. Chatbots, copilots, internal assistants, third-party agents, all of them need to understand location, and none of them speak GIS natively, this is where Eagle has been investing its resources.[/vc_column_text][/vc_column_inner][/vc_row_inner][/vc_column][/vc_row][vc_row][vc_column width=”1/1″][vc_row_inner][vc_column_inner width=”1/1″][vc_column_text]
Why most AI projects stall
In order to be successful, we need to understand why AI Projects stall and what are the common traits.
- Unprepared data: poor metadata and unstructured information stall progress long before the model becomes the limiting factor.
- Architectural fragmentation: AI bolted on as a plugin instead of designed in as a core system component.
- The trust tax: non-deterministic outputs and hallucinations erode confidence faster than good demos build it.
- No lifecycle ownership: AI is not set-and-forget. Models deprecate, behaviour drifts, and pipelines quietly rot.
A good AI strategy exists to design these four failure modes out from the start, not patch them in afterwards.[/vc_column_text][/vc_column_inner][/vc_row_inner][/vc_column][/vc_row][vc_row][vc_column width=”1/1″][vc_column_text]
Four intelligences, one architecture decision
AI agent systems are powerful, but they are not magic. Choosing to use AI involves strict trade -offs across accuracy, cost, complexity, and risk, and the right answer is rarely “just add AI.”
We work from four distinct kinds of intelligence, each strong on different axes and weak on others:
- Humans: unmatched judgement, flexibility, and accountability, but cannot scale. Right for high-stakes decisions, ambiguity, and anything requiring ownership or liability.
- Rules and code: fast, cheap, and fully deterministic, but brittle the moment conditions change. Right for explicit logic: payment checks, validation, access control.
- Machine learning: scales on patterns too complex to write down by hand, but is hard to explain and needs ongoing maintenance as data drifts. Right for fraud detection, forecasting, recommendations.
- Generative AI: flexible with unstructured input and fast to deploy, but non-deterministic by nature. Right for summarisation, document Q&A, and interpreting messy human language.
No single option wins on all four axes, accuracy, flexibility, cost, and determinism, the answer is not picking the right model, these are architectural decisions that should be driven by requirements: combining the four so each does only the part it’s actually good at.
The pattern we build to: a human asks a question in plain language. GenAI translates that intent into a call to deterministic automation the one thing GenAI is genuinely excellent at. Rules and code then do the actual work or the machine learning algorithms, repeatably, the same way every time. GenAI translates the result back into plain language. The human owns the decision.[/vc_column_text][us_separator size=”small”][us_image image=”4500″ align=”center” link=”%7B%22url%22%3A%22%22%7D”][vc_column_text]
GenAI Pattern
[/vc_column_text][us_separator size=”small”][vc_column_text]The non-deterministic part GenAI only ever sits at the edges, doing translation. The answer itself comes from code that can be tested and audited. That’s how we get the flexibility of natural language without paying for it in hallucinations, and it’s the same discipline behind CAS-agent below: the AI decides which approved workflow to run, never what the answer is.[/vc_column_text][/vc_column][/vc_row][vc_row][vc_column width=”1/1″][vc_row_inner][vc_column_inner width=”1/1″][vc_column_text]
Eagle’s Geospatial AI Framework
We’ve built a five-layer framework, wrapped in an aligned strategy and backed by ongoing support, that takes geospatial data from “AI can’t see it” to “AI can trust it”:
- Foundation and Governance: Making metadata AI-ready and wrapping access in the ArcGIS security model, so AI inherits each user’s existing permissions rather than bypassing them.
- Knowledge Bases: Building a secure, searchable index over your spatial data, enabling Retrieval-Augmented Generation (RAG) so answers are grounded in your own information, not guesswork.
- Geospatial Integration: An MCP (Model Context Protocol) server that exposes ArcGIS Enterprise, ArcGIS Online, and the Location Platform to any compliant AI client, secured by token, OAuth, or API key.
- Deterministic Reliability: The AI never runs a workflow or infers intent on its own. Specialised agents select from preconfigured, business-rule-driven workflows — reducing hallucination risk and guaranteeing repeatable results.
- Guardrails and Monitoring: Leverage existing model services to provide cost effective generative AI capabilities. These have the added benefit of providing out of the box tooling for monitoring and providing guardrails.
- Ongoing Lifecycle Support: Because the newest model isn’t always the best one, and every model update changes how the system needs to be tuned, we treat AI as a managed service, not a one-off project.
The unglamorous layers, data readiness and knowledge bases, are where the real value sits and provides a strong foundation on which to build any AI system.[/vc_column_text][/vc_column_inner][/vc_row_inner][/vc_column][/vc_row][vc_row][vc_column width=”1/1″][vc_row_inner][vc_column_inner width=”1/1″][vc_column_text]
Proof in practice: CAS-agent
Frameworks are only as good as what they produce. See the Crash Analysis System (CAS) agent, built by Eagle’s AI engineering team, gives non-specialists a safe, repeatable way to run crash analysis workflows against NZTA’s public Crash Analysis System data without needing to know the underlying layers, filters, or spatial query tools.
Crucially, CAS-agent is not a general chatbot. It’s a controlled coordinator: it reads intent, asks clarifying questions, selects an analyst-approved workflow, and explains the result in plain language. It cannot access data directly, write arbitrary queries, alter records, or invent a metric. Every calculation runs through deterministic, auditable system code the AI’s job stops at interpretation and explanation. This ensures that you will always get the same answer each and eveytime, because the geospatial system is running the workflows, not the model.[/vc_column_text][/vc_column_inner][/vc_row_inner][/vc_column][/vc_row][vc_row][vc_column width=”1/1″][vc_row_inner][vc_column_inner width=”1/1″][vc_column_text]
An investment, not an experiment
Eagle has a dedicated AI Engineer role built specifically to develop and support this framework because doing this well takes more than a chatbot wrapper around a language model. It takes governance, security, deterministic design, and a commitment to keep tuning the system as the underlying models change.
Esri is making GIS smarter. Eagle is making AI smarter about geography safely, deterministically, and built to last well beyond the first demo.
Want to know where your organisation sits on the AI readiness curve? Get in touch with the Eagle Technology Group team to talk through a strategy workshop or readiness assessment.[/vc_column_text][/vc_column_inner][/vc_row_inner][/vc_column][/vc_row]