"They work great in the lab"

Exoskeletons work. That's no longer the question. A device that works in controlled conditions and a device people actually use are not the same problem. The technology press has spent fifteen years failing to notice the difference. We're watching the third cycle of this exact mistake.

In 2011, DARPA's HULC (Human Universal Load Carrier) demonstrated precisely what current reporting celebrates: measurable, noticeable benefits in real-world tasks. Soldiers could carry more weight farther. The device was deemed ready for deployment. By 2015 it was gone, killed not by engineering failure but by soldier rejection. The reason was humbling. A single person in an exoskeleton moves slower than a squad distributing the same weight across multiple bodies. The math of logistics beat the math of individual load capacity.

That lesson — that a technology's lab performance and its adoption in actual workflows are unrelated problems — vanished from the conversation. Current exoskeleton coverage repeats the HULC pattern exactly: controlled-setting benefits, real-world promise. Strategic silence about the barriers that actually determine whether humans adopt a tool. Cost per unit, maintenance requirements, training time, failure modes under fatigue, whether it slows you down in the real task, whether your coworkers now have to work around your device instead of with you.

The companies deploying exoskeletons today are niche industrial and medical operations where the cost and complexity calculus might actually work. Warehouse workers in a few pilot programs. Physical therapists in rehab settings. These aren't false claims. They're incomplete ones. The gap between "measurable benefits in controlled settings" and "this reshapes how people actually work" is where every exoskeleton graveyard sits. Until reporting names that gap, we're just watching the prototype get rebranded as progress every five years.

This isn't unique to exoskeletons. Any technology that requires users to change their actual behavior runs into this wall. The tool that works best in isolation rarely works best in a system where everyone else is still using last year's method. That's not a limitation to overcome with better engineering. It's a permanent feature of how humans coordinate. Missing it. Reporting as though a controlled benefit settles the real question. Mistakes a lab result for an adoption story.