RailX: designing rail orchestration for a future that hasn't shipped yet
An agentic, spatial workbench for US freight rail — built at the World Model Hackathon
This one started with a LinkedIn post.
Chris Haas — a former colleague from Blume, now WiseTech, where he runs business development for rail — put something up, and it pulled a whole stretch of good memories back with it. I loved working on rail. And I have a slightly unfair relationship with the subject, because I came to American freight rail from Europe, and the first honest reaction was: this is the Wild West. The level of fragmentation and improvisation genuinely leaves me speechless sometimes.
Which, for me, is a feature, not a bug. More chaos means more problems worth solving, and more problems means more interesting products to build. So when the World Model Hackathon announced that one of the tracks was agent orchestration through spatial workbenches and immersive dashboards, a "happy marriage" of the five big American Class-I railways was where my head went immediately.
Thinking ten years out
Two conversations at the hack changed the ambition. I got to talk with Greg Madison — a staff XR/AI interaction designer and spatial-computing researcher who was mentoring and judging — and he pushed me to think much bigger about what spatial interaction could do in the most regulated, least-glamorous fields. That's exactly the kind of permission I needed.
So I stopped asking "what could I ship this weekend" and started asking: what does rail agentic orchestration look like ten years from now? And RailX basically appeared in my head.
Here's the version I actually want: I'm a dispatcher, and I have a live model of the train, and I pack it — I can see how every container and load actually fits, cart by cart, box by box. Because in real life the measurements are wrong all the time, something doesn't fit, and dispatchers are rearranging things last minute under pressure. Imagine catching that before the yard does.
That full vision was too futuristic for a hackathon weekend. So I built the thing one step back from it: an agentic, spatially-driven workbench, in the hope that it points at a tool dispatchers might actually get one day. Maybe in five years I dust it off and get to say told you so.:)
Why draggable panels
The interaction I care most about is the customization. Draggable React components were the closest I could get, in a weekend, to that spatial "Lego" feeling — build-a-train, arrange-your-own-cockpit. It mattered to me that the operator could lay the workbench out however they work, not however I decided they should.
That's not a gimmick, it's a bet: I'm a firm believer that agentic-driven UI is going to be deeply, personally customizable, and very soon. The dashboard that rearranges itself to the operator is coming. I wanted RailX to feel like an early postcard from that world.
What it does
Five AI agents, one per major railway, each managing its own network, coordinated by an orchestration layer — with a spatial map that surfaces exceptions in real time. The demo scenario: a blizzard hits a Chicago segment, and you watch the system respond — one agent recalculates a reroute, another pulls back intake, load redistributes across the others, and the whole negotiation streams into an event log line by line. The point of that log is trust: the agents think out loud, and a human stays in the loop to watch and to trigger.
Designing ahead of the build
The best thing Greg said afterward stuck with me: he designs for things that are five years out all the time, and it's our job as designers to be that far ahead. We used to just wait for dev to catch up. These days it's less waiting and more building alongside engineers in the same room — which, honestly, is the fun part.
RailX is live at rail-x.netlify.app. It's a concept. For now.