AI Week Update: Bad (Humanoid) Robot 🤖

🔥

One surprise from AI Week:

I expected more discussion about humanoid robotics.

A lot more.

Instead, it felt like a category sitting just outside the center of the AI conversation… despite potentially having some of the largest long-term implications.

One of the few talks focused on humanoid robotics came from Porsche Consulting. Their perspective was insightful, partly because it lacked some of the hype that surrounds, dare I say it, conventional AI conversations.

The physical robot presence at the event told its own story: a non-working Tesla Bot replica, two robot dogs on leashes, and one robot that did not move.

That was it.

For all the talk about robotics, there was very little actual working robotics on the floor.

And maybe that was the point.

The category is exciting, but still early. The demos are ahead of the operations. The social media clips are ahead of the business use cases.

The backflips are ahead of the pick-and-pack work.

Porsche Consulting highlighted that same point. The robot doing backflips on social media creates a lot of buzz, but that does not mean it can pick and pack simple items inside a real workflow.

They showed a video of that same robot attempting basic pick-and-pack work.

Honestly, it was not impressive at all.

Movement is not the same as operational usefulness. A backflip creates attention. Real value comes from repeatability, accuracy, exception handling, uptime, safety, and cost per completed task.

As a former COO, there was one statement that jolted my thinking:

Humanoid robotics is probabilistic.

These systems operate on confidence levels: identifying objects, understanding environments, making movement decisions, handling exceptions, and deciding whether to continue or escalate.

Robot logic increasingly sounds like this:

“Proceed at 92% confidence or stop?”

As someone who spent years in operations, that statement was hard to ignore.

In traditional operations, no self-respecting process would be designed around a 92% confidence threshold. If a manufacturing line was wrong 8% of the time, it would be considered broken.

But the race to win the humanoid robotics market is forcing a different kind of evaluation.

Not whether the robot is flawless.

Whether the robot can be managed, measured, corrected, and improved faster than the current process.

That is why another point stood out: current humanoid deployments still require human adjustment. The presenter said the average is around four adjustments per day, but added that this is still fewer adjustments than the average human worker requires.

That made me stop and think.

Real operations are already messy.

People need training.People drift.People get tired.People get sick.People take vacations.People miss things.Processes break.Exceptions happen.

So humanoids may not need to be perfect to become useful.

That leads to the bigger issue: these systems need enormous amounts of real-world operational data to improve accuracy.

A lot of companies will not have enough data on their own.

That creates a huge advantage for companies operating at scale: more environments, more edge cases, more feedback, more training loops, and more learning opportunities.

It also explains why companies are racing to deploy early. Whoever gets robots into real environments first may build the largest data advantage later.

Another thing became obvious: if humanoids move into serious industrial adoption, there will likely be shortages everywhere. Robots, compute, energy, manufacturing capacity, infrastructure, deployment expertise, and operational knowledge.

One slide suggested broader adoption could accelerate around 2028+, but the underlying message felt clear:

The strategic positions may be taken long before mass adoption becomes obvious.

And if all of this was not enough already, then came what may have been the quote of the show:

“The robot is the customer. You are the supplier.”

Ok already, let’s stop while we are ahead.

Even if this ends up being one of the smartest ideas in human history, when I say that quote out loud…

…it sounds an awful lot like:

“We serve the robots.”

My biggest takeaway:

Humanoid robotics may still be early, but the direction is hard to ignore.

AI is starting to move from the screen into the physical world.

And when that happens, the winners may not be the companies with the best demo.

They may be the companies with the most data, the best infrastructure, the lowest energy cost, the most reliable systems, and the fastest learning loops.

That is the race underneath the robot. 🤖

7 comments