beyond-gis

The Pattern Transfers: What AI Unlocks Beyond ArcGIS

I thought it was a GIS story.

For the first few months of working this way — describing maps in plain English, getting working ArcGIS SDK code back — I assumed it was specific to Esri's toolkit. Something about the way the SDK was documented, maybe. Something about the particular way AI had absorbed it.

Then I tried D3.

Same result. Complete, working code. First response.

Then Three.js. Then Plotly. Then Python scripts I'd been avoiding for years because the syntax never stuck. Same result every time.

It's not a GIS story. The pattern transfers completely to every complex toolkit that's ever been gated behind years of coding expertise. ArcGIS was just where I discovered it.

The pattern

Before the tools — the pattern itself:

Describe the output, not the code. Get back a complete, deployable file. Run it. Iterate by describing what you want changed.

That's it. Every toolkit below follows this sequence. The only thing that changes is what you're describing.

Data visualization

D3.js is to charts what the ArcGIS SDK is to maps. Thousands of pieces, infinite flexibility, historically specialist-only. I've used it for animated disaster timelines, custom scatter plots, interactive data tables — things that would have taken days to learn from scratch and would have looked worse.

The New York Times graphics desk uses D3. I use it now too, for the same reasons they do: it does exactly what you describe and nothing else. AI eliminated the gap between knowing what you want and being able to build it.

Plotly and ECharts are professional-grade interactive dashboards. Think Power BI or Tableau, deployed as a single file, fully custom, no license, no vendor lock-in. I built a donor analytics dashboard this way. The data was an Excel export. The output was a password-protected Vercel URL with bar charts, filters, and drill-down popups. It took one afternoon. IT would have scheduled that project for Q3.

3D and simulation

Three.js is the WebGL toolkit for 3D graphics in the browser. Ordinarily requires a specialist. I used it for a territory realignment proposal — an interactive 3D diagram executives could rotate in a browser without installing anything. The alternative was a static PowerPoint slide. The output was demonstrably better.

Cesium is specifically 3D geospatial. Think ArcGIS SceneView but open source. Satellite imagery, terrain, 3D buildings, flight paths. NASA uses it. One prompt, one HTML file, one URL that runs.

Documents and reporting

PDF generation, Word documents, Excel files — AI can generate complete professional reports programmatically from live data queries. Not copy-paste. Not mail merge. Actual styled documents with your data baked in.

Situation reports are the obvious Red Cross application. The kind that go to senior leadership at 2am — currently written by a human pulling numbers from three systems while exhausted. A Python script that queries AGOL, formats the data, and outputs a Word document in Red Cross template format is achievable. I've prototyped pieces of it. Getting the full version into production is on my list for this year.

Automation

Python scripting is where I've seen the biggest day-to-day change. I'd been using the ArcGIS Python API for years — well enough to get things done, slowly. Now I describe the transformation, get the complete script, run it. The speed difference isn't 2x. It's closer to 10x on anything complex.

n8n — I have a workflow that syncs Gmail fundraising intelligence to a NotebookLM notebook every morning at 7am. AI writes the custom workflow logic that n8n can't express natively. The ceiling on automation disappears.

Web applications

Any web app. Forms that push directly to AGOL Feature Layers. Chapter locators with custom clustering. Training tools with embedded maps and completion tracking. Internal Red Cross tools that would normally require an IT project become afternoon builds.

The pattern is the same: describe the tool, get the code, host on Vercel, share the URL. IT never gets a ticket.

AI-powered applications

This is the one that compounds everything else.

The application itself calls an AI API. The map I'm running as a live demo shows the baseline — working code from plain English. The next layer is the application thinking alongside the user.

Patterns I've built or have running:

Upload a damage assessment photo → AI analyzes structural damage → structured report
Paste a shelter intake form → AI extracts addresses, geocodes, pushes to AGOL
Ask a question about your AGOL data in plain English → AI queries the Feature Layer and answers

You're not just building maps anymore. You're building tools that combine your domain knowledge with AI reasoning. The output is something neither you nor the AI could have produced alone.

The thing nobody says out loud

Every toolkit that was previously gated by coding expertise is now accessible. Not just ArcGIS. Not just maps. All of it.

The reason this matters specifically for GIS professionals: you already have something most developers don't. Deep domain knowledge. You know what your organization actually needs. What data exists. What questions matter at 2am during a disaster response. What a field team needs on a device with bad cell service.

Developers can write code but don't know the domain. You know the domain and can now write the code. That combination is rare. Genuinely rare. And genuinely valuable.

What to actually do

Don't try to learn all of it. A sequence that works:

Next 30 days: Stay in your current groove. Speed compounds faster than breadth.

Next 90 days: Pick one adjacent toolkit. D3 or Plotly is the natural next step — your data stories deserve better charts than AGOL Dashboards produce. One working visualization changes your expectations permanently.

Next 6 months: Build one AI-powered tool. Something that takes data in, runs it through an AI API, and produces something useful out the other side. One working demo changes how leadership thinks about what's possible.

The long game: you're not a GIS developer who learned to use AI. You're building toward being a domain expert who can prototype any tool any team needs — faster than IT can schedule a discovery call.

That's a different professional identity than the one you started with.

The pattern is the skill. The SDK was just where I learned it.