The Hype-vs-Reality Gap: How Much of What You See in Demos Is Real?

A healthy amount of skepticism is the correct default setting for this entire field, and that's not an insult to it.

Nearly every humanoid robotics demo video generates the same split reaction online: genuine excitement, and a chorus of "but is it really doing that on its own?" Both reactions are usually at least partly justified.

What's actually going on behind most demos

A lot of what reads as full autonomy in public footage is closer to a mix of scripted state machines for the well-rehearsed parts, teleoperation for the genuinely hard sub-tasks, and real autonomous execution for whatever's left. Transparency about that mix varies enormously between companies, and very few disclose it in detail.

That's not necessarily dishonest — a demo is meant to show a capability is possible, not that it's production-ready — but it does mean the gap between "demoed once" and "reliably autonomous for an eight-hour shift, every shift" is usually much larger than the video suggests.

A useful pattern to watch for

A recurring pattern in this field: a capability looks solved in a controlled demo, then takes another 18 to 24 months of unglamorous engineering — edge cases, failure recovery, consistency — before it's boringly reliable enough to matter commercially. Figure's eleven months of real BMW production hours is valuable precisely because it's the opposite of a one-off demo.

How to read a claim more skeptically

Ask whether a number is a shipped-and-verified figure or an announced production target — those get conflated constantly. Ask whether "deployed" means one unit in one facility or a real fleet across multiple sites. And treat any company's self-reported uptime or reliability statistics as provisional until independent, third-party data exists to check them against — which, for most of this industry in 2026, it still doesn't.