The spatial intelligence stack: capture, understand, act

“Spatial intelligence” sounds abstract until you break it into layers. Once you do, it becomes a checklist you can actually build against — and a way to spot which layer a given product is really competing on. I think of it as three stacked jobs: capture, understand, and act.

Layer 1 — Capture

Before software can do anything with space, it has to sense it. A decade ago this needed specialist rigs; now any modern phone can scan a room, and lidar, depth cameras, and photogrammetry fill in the rest. The output is raw: a point cloud, a mesh, a stream of frames. On its own it’s just geometry — useful, but dumb. Capture is increasingly a solved, commodity layer. Competing here alone is a race to the bottom.

Layer 2 — Understand

This is where it gets interesting. Understanding turns raw geometry into meaning: that surface is a floor, that shape is a chair, that flow of movement means people avoid this corner. It’s the difference between a scan of a room and a system that knows it’s a room. This is also where behaviour lives — gaze, dwell, hesitation — and where most of the durable value sits, because interpretation is hard and compounding.

Layer 3 — Act

Understanding is wasted if nothing happens with it. The action layer is where intelligence touches a person or a decision: anchoring a digital object to a real surface, surfacing the right information at the right place, or feeding a recommendation back to a planner or designer. Good action feels effortless and almost invisible — the answer is simply there, in the space, when you need it.

Most “AR apps” only do capture and a thin slice of action. The winners go deep on understanding — that’s the moat.

Why the model is useful

When you’re evaluating a product, an idea, or a vendor, ask which layer it actually competes on. A clever overlay with no real understanding is fragile. A system that genuinely understands space and behaviour — like a VR space audit reading emotional response, or AR crowdsourcing turning placements into planning data — has something defensible. And tools like Magic show the same logic in content: the value isn’t the render, it’s the understanding that makes one button enough.

If you want the bigger picture of why this matters now, start with what spatial intelligence is.