While Humanoid Robots Dance at Expos, Do We Really Need Them to Work?

At the WAIC 2026 venue, humanoid robots are undoubtedly the brightest stars. Dancing, fighting, running, playing table tennis—actions that were once only seen in labs have become staple shows on exhibition stands. But one booth was a bit different—the robots were seriously working.

As a cross-border e-commerce operator, my daily work revolves around product selection and copywriting. Over the past two years, I've used AI tools for these tasks, starting with ChatGPT for writing copy, then using Stable Diffusion to generate product images, and later using Python scripts to scrape competitor data. This workflow has saved me quite a bit of time and slightly improved actual conversion rates. But I know these tools are useful because they solve specific problems—not showing off, but genuinely helping me get work done.

So when I saw those dancing robots at WAIC, my first reaction wasn't excitement, but confusion. No matter how well these robots dance, can they help me move goods in the warehouse? Can they help me pack and ship orders? If it's just performance, what's the difference between them and the dancing AI virtual humans I see at other expos? It's just a change of shell.

But one company's booth made me stop scrolling. The news said their robots are seriously working—not performing, but doing actual jobs. For example, moving materials in simulated warehouse scenarios or completing assembly actions on production lines. Although I'm not sure exactly which company it is, this direction resonates with me.

I understand that showcasing muscle and athletic ability is part of technical validation, just like when I first learned to code, I would write some cool-looking crawlers to scrape public data, but what actually generated value were the scripts that ran stably and automatically handled abnormal data. Robot development should follow the same path—from lab to factory, from performance to working—that is true implementation.

[!note] I noticed a detail: many robot companies showcase high-difficulty moves in their promotions, such as backflips and parkour. These moves do attract attention, but the underlying engineering capability isn't necessarily stronger than robots that can stably execute repetitive tasks. Just like my AI copywriting tool, being able to write poetry and being able to write product descriptions are two different things.

My own experience is that when using AI tools for product selection, the biggest headache isn't that the model isn't strong enough, but how to make the model understand my specific needs. For example, I wrote a simple prompt template:

You are a cross-border e-commerce operator, need to write a product description for a [Product Name].
Target market: [US/Europe/Japan]
Product features: [List 3-5 core selling points]
Target users: [Age group Gender Interests/Hobbies]
Writing style: Concise, persuasive, emphasizing use cases rather than functional parameters.
Output format: Title (within 30 characters) + 5 bullet points + a summary paragraph (within 80 characters)

This template looks simple, but in practice, the output quality is much higher than when I frequently adjusted prompts before. The reason lies in solving real problems—how to quickly generate product copy that complies with platform standards. Similarly, if a robot can stably complete moving, sorting, and assembling in a factory, even if the movements don't look that cool, it is more valuable than a robot that can dance but can't work.

I noticed the news mentioned that the robot movements at this company's booth looked somewhat "clumsy"—not smooth

Original link: https://www.qbitai.com/2026/07/458348.html