The gap between the demo and the deployment

Every robotics pilot I have read about follows the same arc. The demo is flawless. The pilot is promising. The deployment is — let's call it a learning experience.

The machine didn't get worse between those stages. The inputs did.

What the demo runs on

A demo runs on the best version of reality. Clean, uniform, rigid boxes. Even lighting. A conveyor that never jams. Items presented one at a time, face up, exactly where the vision system expects them. Every variable that can be controlled, is.

That is not cheating. It is how you prove a mechanism works at all. But it quietly assumes the hardest part of the job — the inputs — away.

What the deployment runs on

A vision system that reads 99.5% of clean labels can fall off a cliff when the label arrives crumpled.

Production runs on whatever the trailer gives you. Crushed corners. Slumped poly mailers. Labels that creased somewhere three states ago. A barcode that is technically present and functionally unreadable. The machine is identical; the world feeding it is not.

That is the gap. And it is not a rounding error — it is the whole story:

# the demo                  # the deployment
input: clean_box            input: whatever_the_trailer_gives_you

Why I keep coming back to this

That gap — between what automation promises and what it delivers on a live floor — looks small in a demo and enormous on a Tuesday night. My hunch is that it matters more than the headlines let on: investments, jobs, which companies hold up over the next decade. I'm not claiming to have that figured out. I'm saying it's the thing I can't stop noticing.

I lead operations next to these machines every shift. This is the starting point — the first entry in a site that's me teaching myself the engineering underneath them in public, and writing down what the gap actually looks like from the deployment seat. Public sources, nothing that names where I work.

From the real world.