The Spacing in Robot Demos Is Off

Watching the drama of Mech-Mind's founder blasting Galaxy General, what I focus on is the gap. In design systems, the scariest thing is mistaking demo states for production states. Videos showing nut-cracking, picking up glass shards, and folding clothes look smooth, leading audiences to easily interpret them as ready-to-use. But people in the embodied intelligence industry know that grasping irregular shapes and soft objects is inherently difficult. Difficulty isn't the problem. Difficulty being edited into pretty short clips is the problem.

When writing component documentation, I fear most writing only "supported" without writing boundaries. Whether a component works depends on empty states, error states, and extreme inputs. Robots are the same. Showing one successful action is like animating a button—it looks good but doesn't prove interface stability. There should be a scenario matching card clarifying objects, materials, failure rates, and how much human takeover is needed. Interaction flows can be optimized, but the premise is admitting where things break. Without this, valuation relies entirely on camera language.

So I don't think this is just mud-slinging. It's more like the industry hasn't widened the distance between description and action. Mixing instructions into descriptions in llms.txt causes issues; mixing demo actions into delivery capabilities in robot marketing causes issues too. One viewer thinks it's usable, one buyer thinks it's deliverable, with dozens of on-site debugging sessions missing in between.

The next round of competition in embodied intelligence will depend on who designs failure states more honestly. Demos can be filmed smoothly, but products must be verifiable. If the gap isn't widened, delivery will face problems later.