Lights-Out Lessons: Why Auto Plants Prove We Don’t Need Humanoid Robots

🏭 Lights-Out Lessons: Why Auto Plants Prove We Don’t Need Humanoid Robots

In 2014, General Motors gave me a tour of the giant Lake Orion plant North of Detroit. Because of automation, the facility was using only about a quarter of the space it was originally built for. Factories have gotten dramatically smaller over time primarily because automated, specialized robots and machinery can operate within much smaller working envelopes and require significantly less aisle space and safety clearances than human workers. The remaining footprint at the Orion plant was divided into four highly efficient quadrants.

The first quadrant, where they weld cars together, had zero human workers. None. It was a sterile, self-operating environment that genuinely looked like a scene from the Terminator movie.

In the second quadrant, cars were painted. There were only a couple of people present, primarily to monitor and fix robots. This minimal human presence was mostly due to the hassle of entering and exiting the clean room environment.

In the third quadrant, specialized robots were automatically charging in the corner. When needed, they would roll out to install the engines from the bottom and bolt them into place. It was incredible to see this choreography in action.

The fourth quadrant was the only area where a few employees were stationed, mainly to put in the center consoles and some smaller interior pieces. All of the necessary tools hung directly from the line and moved along with the cars. The rode on a belt that moved underneath it all. The entire process was efficient beyond anything I had ever witnessed.

 

The Automation Roadmap

I asked my sponsor at GM why those few employees in Quadrant 4 still existed. He informed me that they were only there because, for now, they were cheaper than the specialized automation required to replace them. However, he confirmed they had a multi-year roadmap to replace every single worker as the cost of the necessary robots came down.

This trajectory is not unique. A couple of years later, Skoda built the first truly “lights-out” car factory. Around the same time, Swatch Watches built the first automatic watch entirely in a lights-out factory with zero human input—even the final tuning was done automatically, eliminating the need for a master watchmaker. This was particularly astounding given that watchmaking is one of the oldest and most revered crafts, a tradition spanning centuries that has historically demanded decades of highly precise, meticulous labor from trained artisans. But not a single humanoid robot in sight….

 

Dynamic Customization

The craziest part of the GM plant was the complexity they were handling. They were building three completely different car models on the same line, and every single car was customized based on specific customer orders from dealerships. They could even dynamically change which customer order a specific chassis filled, as long as the car hadn’t yet been configured with a conflicting, irreversible feature. For example, if a customer at one dealership paid a premium for a specific option, and a similar car was half-finished on the line, they could dynamically divert and convert a car further down the line into that customer’s exact vehicle. This level of dynamic manufacturing was incredible. They had essentially perfected this system by 2011, and by 2014, it was running smoothly, handling software updates and model changes with ease.

 

The Myth of the Humanoid Robot

Years later, Elon Musk famously said, “it turns out, building the cars is way harder than designing them. I didn’t know.” That statement is idiotic. All he had to do was tour a modern car plant like Orion and have them explain how it works, just as they did with me. He literally did zero homework, or he was lying. None of the highly effective robots in car plants are humanoid. None in the Swatch plant are. They don’t need to be. Specialized, non-humanoid machines are simply cheaper, easier to maintain, and fit into the required physical spaces better. Again, Musk should know this after managing car plants for many years now. He’s either lying, or he’s an idiot.

---

Separately, the energy paradox poses a significant hurdle for creating highly capable, autonomous humanoid robots, particularly those that need to run complex, modern AI models. These models, which often rely on powerful graphics processing units (GPUs) for efficient inference, demand substantial electrical power. However, robots designed for mobility and human-scale interaction are often too small to carry batteries large enough to sustain these power-hungry GPUs for extended periods. The common fallback solution—offloading the intense computational workload by connecting the robot to the internet and using remote GPUs in a data center for inference—introduces an even greater failure point. This dependency on external connectivity becomes a critical vulnerability when the robot, such as a hypothetical plumber robot, operates in environments with poor or nonexistent network access, like a deep basement or inside thick walls, rendering it effectively “stuck” and incapable of making real-time decisions. I wrote more about it here: The Ballad of Pipe-Bot 5000: Why Humanoid Plumbers are Just Heavy Paperweights.

---