00 / Opening
AI, GIS, and the new Red Cross workbench.
Not a prompt trick. A working model for maps, memory, data, prototypes, and decision support.
01 / The shift
The story is bigger than prompt to app.
The obvious win is GitHub to Vercel. The durable win is a full workbench: project recall, map logic, data cleanup, deployment, documentation, and review.
AI does not replace GIS judgment. It gives GIS judgment a faster workbench.
02 / Hidden superpower
A folder becomes a memory capsule.
The project keeps its decisions, errors, fixes, source files, commits, notes, screenshots, prompts, and next actions. Six months later, the work is not gone.
- Obsidian is the readable memory layer.
- GitHub is the version and accountability layer.
- Vercel is the visible proof layer.
/notes /source-data /app /screenshots AGENTS.md handoff-2026-05-19.md
Ask: what is this? what broke? what next?
03 / Prompt to public app
A URL changes the conversation.
A prompt can become a repo. The repo can become a live page. The live page can become a demo, a leadership artifact, or an Experience Builder embed.
When it has a URL, it is no longer a private experiment.
04 / ArcGIS acceleration
AI meets the real ESRI workbench.
The high-value work is not abstract coding. It is Arcade, popups, Experience Builder embeds, SDK prototypes, schema cleanup, layer filters, and notebook outputs.
- Write rich Map Viewer popup cards with Arcade content dictionaries.
- Turn map requests into testable ArcGIS SDK apps.
- Translate plain-English intent into fields, filters, code, and UI behavior.
return {
type: "text",
text: "<b>" + $feature.NAME + "</b>"
+ "<br>Chapter: " + $feature.CHAPTER
+ "<br>B&O: " + $feature.BO
};
05 / Backstage frontstage
Notebooks backstage. Maps and apps frontstage.
Python and ArcGIS notebooks are strongest as automation, QA, exports, and repeatable analysis. Polished experiences belong in HTML, GitHub, Vercel, Map Viewer, and Experience Builder.
Use the right surface for the audience.
06 / Red Cross strategy support
AI helps assemble evidence so people can decide.
The strategic tools are not final answers. They are evidence surfaces for humans: county scenarios, real estate context, donor geography, facility use, vulnerability data, and operational history.
07 / Disaster intelligence
What if the map could explain what changed?
SitAware, Cascade, smart-query, and anticipatory briefings point toward a new operating pattern: maps that do more than display layers.
- Live hazards and ArcGIS overlays.
- Flood impact AOIs and chapter/region context.
- Natural-language questions against county and disaster data.
08 / Knowledge systems
Search finds notes. RAG finds implications.
Ask Clara, Dragons Brain, LightRAG, and county intelligence show a different pattern: source-grounded answers across policy, local notes, federal data, Red Cross hierarchy, and county context.
The archive becomes an active work surface.
09 / Data pipelines
From spreadsheet chaos to managed workflow.
Volunteer data, AGOL audits, metadata cleanup, and CSV-to-ArcGIS workflows show AI's practical value: repeatability, edge-case checks, and documented process.
The script matters. The repeatable process matters more.
10 / Training
Teach one useful workflow at a time.
The EB Teaching App and embed curriculum point to a practical enablement strategy: make ArcGIS concepts plain enough that people can use them immediately.
- Map, layer, view, filter, Arcade, SDK, AI briefing.
- Concrete examples beat abstract training.
- The presentation can become a reusable training artifact.
Map = where things are Layer = what you are showing View = what the audience sees Filter = which records matter Arcade = how the popup thinks SDK = when the app needs more
11 / Human-centered prototypes
Some ideas need to be seen before they can be judged.
Language access prototypes, multilingual RAG, and emergency-app mirrors make policy ideas visible as interactions. They do not need to be final products to move the conversation forward.
A realistic demo can make a hard idea concrete.
12 / Governance
Inaction is also a choice.
AI risk is real. So is the risk of waiting while existing systems keep producing errors, delays, stale data, and missed opportunities.
The comparison is not AI versus perfection. It is AI-assisted work versus current reality.
13 / Agentic workflows
Small trusted workflows beat one giant agent.
Skills, AGENTS.md files, project rules, and local instructions turn generic AI into repeatable project behavior.
- Raw memory to cleaned notes to useful outputs.
- Prototype-to-Vercel when the goal is a live proof link.
- Parallelize exploration. Serialize deployment.
skills/ memory/raw/ memory/wikis/ memory/outputs/ automations/ Use the skill. Save the evidence. Verify the result.
14 / Learning and confidence
The fear cost of trying goes way down.
For a GIS professional learning fast, AI changes the emotional math. Advanced coding, deployment, debugging, and design become inspectable and repairable.
Not "become a software engineer." More like: build enough to make the GIS idea real.
15 / Proof objects
Make the invisible workflow visible.
The presentation should collect proof, not just claims. Screenshots, URLs, notebooks, popup code, deployment logs, source traces, and project notes make the story concrete.
16 / The ask
Treat this as a practical capability to learn and share.
Start small. Pick real GIS workflows. Build proof artifacts. Verify them. Document the path. Teach the repeatable parts.