XPeng put a humanoid robot on stage in November and half the internet lost its mind. The thing walked so smoothly that people accused the company of hiding a person inside a suit, so the CEO had it unzipped on camera to prove otherwise. It's called IRON. It's 178cm and 70kg, it runs on XPeng's own AI chips, and the coverage that followed reads like a countdown to mass unemployment. A robot that can walk, see, and use its hands. Coming to a factory near you. Better start updating the CV.
I want to push back on that. Not because the technology is fake, but because the "takes all the jobs" story is one we've been told over and over, and it keeps not happening on schedule.
Start with what IRON is actually going to do. Not "replace the workforce." XPeng plans to build more than 1,000 units a month by the end of 2026, with the first commercial deployments in 2027, and those first robots go into XPeng's own stores as sales assistants and tour guides. CEO He Xiaopeng has said plainly that Chinese factory labour is still too cheap to make industrial deployment worth it right now. So the machine built to take every job is starting its career as a shop greeter. That's not a dig at XPeng. It's a useful reality check against the headline.
Now the part that gets me. We have run this exact experiment before, with people far more credible than a product launch.
In 2016, Geoffrey Hinton, the man who later won a Nobel Prize for the neural network work sitting underneath all of this, told a room that we should stop training radiologists immediately. "It's just completely obvious that within five years deep learning is going to do better than radiologists," he said. Reasonable-sounding. Confident. Wrong. Nine years on, radiology is facing one of the worst labour shortages in its history. The Mayo Clinic grew its radiology staff by 55% over that stretch. The American College of Radiology is forecasting the specialty will grow another 26% over the next three decades. AI did arrive in radiology. It turned out to be a tool radiologists use, not a replacement for them. Hinton himself later admitted he got the timing wrong and had really only been talking about image analysis, not the whole job.
Self-driving cars are the other one. "Full autonomy next year" has been an annual tradition for about a decade, and I still can't summon a robotaxi in most cities.
Here's the pattern I'd ask any executive to sit with. A demo is not a deployment. A robot walking beautifully across a stage tells you almost nothing about whether it can hold down an eight-hour shift in a messy warehouse, handle the one case in fifty that goes sideways, and still cost less than the person doing that work today. Impressive and reliable are different problems, and the second one is where most of these predictions quietly die. The last 10% of reliability is usually harder and pricier than the first 90% put together.
None of this means IRON is vapourware, or that automation won't change how we work. It will. But it changes work the boring way: narrow tasks, specific settings, years of unglamorous integration, humans still in the loop cleaning up the edges. The jobs that shift first are the predictable, repetitive ones, and even those move slower than the keynote implies. We saw the same pattern play out when economists published their open letter warning about AI's pace: the urgency was real, but the specifics were messier than the headline suggested.
Which is the point I actually care about. If you run a business, the worst thing you can do with a story like this is either panic or wave it off. Both are guesses dressed up as strategy. The companies getting real value out of AI aren't the ones restructuring around a robot they saw on stage. They're the ones automating one specific process at a time, checking whether it actually worked, and keeping their people pointed at the parts machines still fumble.
That's the work we do at OrionX. We're not here to sell you a robot future. We help you find the specific, unglamorous places where automation pays off now, and skip the ones that only pay off in a press release. If you want to talk through where AI genuinely fits in your operation, we're around.
Sources
- XPeng IRON mass production and rollout timeline (1,000+ units/month by end of 2026, commercial deployment 2027, showroom sales-assistant roles first): eWeek, Interesting Engineering, IndexBox
- He Xiaopeng on factory labour costs making near-term industrial/household deployment impractical: MEXC / CoinCentral
- IRON stage demo and CEO revealing internal structure after "is there a human inside?" reactions: Interesting Engineering
- Geoffrey Hinton's 2016 "stop training radiologists" prediction: The New Republic, UAB Reporter
- What actually happened to radiology (Mayo Clinic +55% staff, ACR forecast +26% over 30 years, Hinton's later walk-back): Radiology Business
