AI Isn’t the Product. The Workflow Is.

Every few days someone sends me a new demo. AI generates a house. AI designs a backyard. AI produces a digital twin from a handful of photos. The video is always beautiful. A slow flythrough. Perfect lighting. A slider that moves from “before” to “after” as if complexity has finally surrendered.

What are we doing?

Then comes the question: “Isn’t this what you’re building?”

No.

What they’re showing is a moment. What actually matters is a system.

We’ve been here before. A decade ago everything was “cloud-native.” Before that it was “big data.” The adjective changes, but the pattern doesn’t. We confuse a capability with a product. We assume that because something can generate output, it has solved the harder problem of producing outcomes.

In spatial systems, imagination has never been the bottleneck. GIS has been generating maps since long before AI. CAD has been generating geometry for decades. BIM systems like Autodesk BIM have been coordinating multi-disciplinary models long before anyone typed a prompt into a chat window. The hard part has never been producing something that looks plausible.

The hard part is making it survive contact with reality.

Reality has permits. Reality has budget constraints. Reality has grading plans and material lead times and coordinate systems that disagree with each other in subtle, expensive ways. Anyone who has worked seriously with PostGIS or tried to move data between platforms like Esri ArcGIS and a construction workflow knows that the “pretty output” is just the beginning of the story.

AI collapses the distance between idea and visualization. That is powerful. But visualization is the visible ten percent of the problem. The invisible ninety percent is everything that happens after someone says, “Okay, now ship it.”

That invisible layer is workflow.

Workflow is not cinematic. It does not make for a compelling product demo. Workflow is where authentication is enforced, where state is tracked across revisions, where metadata stops being decorative and becomes operational. It is where assumptions are made explicit instead of silently embedded in someone’s head. It is where integrations either behave deterministically or quietly corrupt downstream systems.

If you’ve ever looked at how something like STAC evolved, you’ve seen this tension in action. STAC didn’t succeed because it was flashy. It succeeded because it functioned as a contract — a shared understanding of structure that allowed systems to interoperate. The same is true for cloud-optimized formats like Cloud Optimized GeoTIFF (COG). The format is not the product; the discipline around how it’s used is what enables scale.

AI-first startups tend to learn this the hard way. They build an extraordinary renderer. They generate layouts, scenes, suggestions. And then a customer asks a devastatingly simple question: “Can this integrate with my existing workflow?”

That question is where demos go to die.

Integration is not just about APIs. It’s about guarantees. It’s about versioning. It’s about knowing that when a field is renamed, something downstream doesn’t silently misinterpret it. It’s about being able to audit who changed what and when. It’s about being able to run the same process twice and get the same answer. That’s not glamour. That’s infrastructure.

I’ve written before that humans are not a scalable integration pattern. AI does not magically change that. If anything, it makes the problem sharper. The faster you can generate possibilities, the more disciplined you have to be about which possibilities become commitments. Without structured workflow, AI simply increases the volume of ungoverned output.

And here’s the part that makes people uncomfortable: AI models are becoming commodities. The differentiation window is shrinking. Today’s breakthrough model is tomorrow’s open-source checkpoint. If your competitive advantage is “we call a model,” you do not have a moat. You have momentum.

Real defensibility accumulates elsewhere. It accumulates in encoded business rules. It accumulates in metadata that is mandatory rather than optional. It accumulates in the connective tissue that moves data from ideation to execution without relying on a human to reconcile inconsistencies at 10pm on a Sunday.

Industries are not transformed by prettier renderings. They are transformed by better coordination. By systems that make assumptions visible. By workflows that survive scale. By contracts — technical and operational — that reduce ambiguity instead of amplifying it.

AI will absolutely reshape spatial computing. It will compress ideation. It will suggest alternatives humans might never consider. It will accelerate the early phases of design in ways that feel like magic.

But magic is not a product.

The product is the system that takes that output, constrains it, validates it, versions it, prices it, and makes it buildable. The product is the workflow that connects imagination to accountability.

The rest is just a very good video.

Powerpoint is boring