On Thursday, Figure AI posted a video that went viral within hours. Two of its F.03 humanoid robots, powered by the company's Helix AI system, cleaned a bedroom and made a bed in under two minutes. Fully autonomous. No teleoperation. The clip racked up millions of views and reignited the dream of general-purpose household robots becoming reality.

The same day, Bloomberg published a deep investigation with a very different message: "Humanoid Robots Aren't as Advanced as the AI Hype Cycle Suggests." The machines can jump, dance, and go viral, Bloomberg reported, but turning them into useful workers remains far more difficult and expensive than their boosters suggest.

Two stories. Same day. Completely opposite conclusions. Welcome to the humanoid robot debate in 2026.

The Demo That Went Viral

Figure AI's demonstration is genuinely impressive on the surface. Two bipedal robots working together, coordinating their movements to strip a bed, fold sheets, fluff pillows, and arrange everything neatly. All in under 120 seconds. The company's Helix system, which serves as the AI brain for its robots, was recently upgraded to what Figure calls its most advanced full-body control system to date.

CEO Brett Adcock is betting the company's future on "general-purpose humanoids" that can eventually handle everything from warehouse logistics to household chores. The bed-making demo is meant to prove the concept: robots that can manipulate soft, deformable objects (sheets, pillowcases) in unstructured environments (a messy bedroom) without pre-programmed motions for every possible scenario.

The Reality Check

Bloomberg's investigation tells the other side. The humanoid robot industry has attracted tens of billions in investment capital on the promise that these machines will replace human workers in factories, warehouses, and homes. But the gap between demo videos and deployable products remains enormous.

The core problem: demos are controlled environments. A bedroom set up for a video is not the same as a random bedroom in a random house with a random arrangement of objects. Real-world deployment requires robots that can handle infinite variations in lighting, surface textures, object placement, unexpected obstacles, and the dozens of edge cases that humans navigate without thinking. Making a bed in a lab is engineering. Making a bed in 10,000 different homes is a product.

Then there is the cost question. These robots are not cheap. Figure AI's machines are not priced for consumer markets, and the economics of deploying them as workers still do not pencil out against human labour for most tasks. Bloomberg's sources suggest the industry is years away from the kind of unit costs that would make humanoid robots viable as a replacement workforce at scale.

The Bigger Picture

This tension is not unique to Figure AI. The entire humanoid robot sector lives in the gap between compelling demonstrations and commercial viability. Companies like Tesla (Optimus), Boston Dynamics, Unitree, and Sanctuary AI are all racing to be first to market with a useful humanoid. China, according to Morgan Stanley, has taken an early lead in the sector, with companies like Leju and Unitree pushing production costs down aggressively.

Siemens executive Rahul Garg offered a middle ground in a podcast this week: "In the next five years, we are certainly going to move from the hype to the scalable commercial deployment." Five years. Not five months. That is the honest timeline from someone inside the industry.

Figure AI's bed-making demo is real progress. It shows meaningful advances in dexterous manipulation and multi-robot coordination. But progress and product-market fit are two different things. And until someone ships a humanoid that works reliably in the real world at a price point the market can absorb, the viral videos will keep coming, and so will the scepticism.