Last week, a Chinese robot claimed it can run faster than Usain Bolt.

Unitree teased a new humanoid it named "Superman" and said it hits 12.66 metres per second vs Bolt's peak was around 12.2.

2 days later Unitree went public in Shanghai, and the stock opened up 629%.

Almost nobody noticed a tiny problem: the robot in that clip has no working hands.

It can supposedly out-sprint the fastest human who ever lived, and it can't pick up a mug. That gap, between what these robots show off and what they can do the second the camera turns off, is the whole story today.

So hold that thought for the next 5 minutes. First, some catchup:

Your coding AI is really 2 things sold together, a brain that thinks and a body that does the work.

This kit shows you how to keep the body and swap in free models for the brain, so you stop paying for the part you can get for nothing.

One key pools dozens of free providers behind a single address on your own machine. You copy-paste the install commands, drop in one config block, and the paid model slots fill with free ones.

It closes with the exact one-shot prompt that built a full Mario-style platformer, Sprout Kingdom, on the free setup.

Unitree's Superman clip is just the newest flex.

This week Beijing runs its second World Humanoid Robot Games, 2,000+ robots, and the organisers banned remote controls this year to force the machines to run on their own.

A lot of the jaw-dropping footage you've seen is either puppeted by a human off-screen or rehearsed into the ground.

One of Tesla's big Optimus bartender demos? A guy in another room was working the controls, and Optimus still can't tell anyone a launch date or a price, a full year into "it's coming".

Making a robot look incredible on camera is easy.

Getting one to figure out what to do by itself, in a room it's never seen is the hard part, and it's the bit the highlight reels skip.

Back in May, Figure put its shiny Figure 03 humanoid in a 10 hour sorting race against one of its own human interns. Packages flying, side by side, all shift.

The intern won, all while he was taking the breaks the robot never needed.

Somewhere a robotics investor felt a small chill.

Why did a person win at boring repetitive sorting, the thing robots are meant to crush?

One word: hands.

Rodney Brooks, who co-founded iRobot, reckons the whole industry is kidding itself here.

Your hand has roughly 17,000 touch sensors in it. He calls the belief that robots will match that any time soon "pure fantasy thinking".

Running fast turns out to be the easy part. The squishy, weirdly shaped, don't-crush-it stuff is where robots still fumble, and that's what most jobs are made of.

While everyone's filming the humanoids, the robots earning their keep look like nothing at all.

Amazon just crossed 1 million robots in its warehouses, and reckons about 75% of its deliveries now pass through a robot somewhere along the way.

Not one of them looks human. They roll around shoving boxes, and they print money.

China's UBTech has its Walker S2 on real BYD and Geely car lines right now, plodding at maybe a third to half of human speed.

Sounds rubbish, right? But that's exactly how every winning automation starts.

Cheap and unglamorous, right up until it's suddenly everywhere.

The most important robot news this year is the one nobody clips for Instagram, and it has no robot body at all.

In late July, DeepMind dropped Gemini Robotics 2, basically an AI brain for robots that handles balance and using both hands together.

Forget building a robot. DeepMind wants to be the brain that every other company's robot runs on.

Same bet at Physical Intelligence (worth around $11B) and Skild AI (over $14B): one AI model that can run almost any robot body, the way one operating system runs on a hundred different phones.

The bodies get all the applause, while the brain decides whether any of this works, and they're improving way faster than the metal is.

One number for the road.

Investors threw $47.4 billion into physical AI in just the first half of 2026, nearly 4x the previous 6 months.

Figure, the top Western humanoid startup, is valued at $39 billion.

Now the fun part.

Goldman Sachs reckons the entire humanoid robot market will be worth about $38 billion, and that's their guess for 2035.

So one company today is priced above the whole market's forecast for 10 years out. At that point the word "valuation" starts to feel generous.

Somebody here is going to be very wrong.

I don't think humanoids are a scam. I think they're about 3 years of demos ahead of where the robots are.

The tell is always hands.

We nailed fast and strong years ago, but nobody’s cracked the human hand, with its thousand tiny thoughtless adjustments to hold a coffee cup without crushing it.

A robot can beat Usain Bolt in a sprint and still lose a fight with a folded t-shirt.

So I read robot videos differently now.

A backflip is an advert, a dull wheeled bot dragging a million parcels across an Amazon warehouse is the actual business, even though nobody's talking about it.

And the money agrees with the boring take.

When a single company is worth more than the entire market is forecast to reach in a decade, "valuation" is doing a lot of heavy lifting.

If you want to know where robotics is heading, ignore the somersaults and watch 2 things: the AI brains coming out of DeepMind / Physical Intelligence, and which companies start posting revenue instead of another highlight reel.

The robot videos got unbelievable in 2026 while the robots themselves still can't fold your laundry. Give it a few years though.

If you read one primary source on this, make it DeepMind's Gemini Robotics 2 post.

And hit reply with the a robot demo that convinced you.

Until next time,
Vaibhav 🤝🏻

PS: the Beijing Humanoid Games kick off on the 22nd. Meanwhile, check this out:

If you read till here, you might find this interesting

#AD 1

When you type a prompt, you summarize. When you speak one, you explain. Wispr Flow captures your full reasoning — constraints, edge cases, examples, tone — and turns it into clean, structured text you paste into ChatGPT, Claude, or any AI tool. The difference shows up immediately. More context in, fewer follow-ups out.

89% of messages sent with zero edits. Used by teams at OpenAI, Vercel, and Clay. Try Wispr Flow free — works on Mac, Windows, and iPhone.

Start flowing free

#AD 2

Running a CX team used to mean following playbooks. AI changed that overnight. Hear three leaders who shaped what came next, Kayla Arp, Chris Rule, and Amy Harvey, on what changed, the challenges they faced, and where CX is headed. Watch on-demand now.

Watch now