I lived in a building with a “smart” delivery robot. It could autonomously navigate hallways, summon elevators, and map dynamic environments using advanced sensors. It couldn’t press a doorbell. It would sit outside my door, wait in silence, and eventually leave with my food. I’d wait an hour, only to find my dinner had auto-returned to the lobby.
We are building trillion-dollar AI brains inside metal bodies that are functionally less useful than a teenager with a bicycle.
You’ve seen the hype. Humanoid robots doing backflips. Embodied AI promising to replace human labor across every industry. But if you’re a Product Manager in this space, you’re trapped in a purgatory of your own making. The industry has completely failed to establish a PM value loop.
I recently went to the World Robotics Conference. I have a stake in a BBQ restaurant, and I asked several robot manufacturers if they could build a machine to replace the guys who manually skewer the meat. They all gave me the same optimistic estimate: 5 to 10 years. I then asked a senior embodied AI expert at a major tech company. He didn’t give me a timeline. He asked, “Are you trying to increase efficiency or attract foot traffic?” I said efficiency. He laughed and said, “A robot can’t do it. But a dedicated automated skewer machine can.” When I asked if the automated machine could grab raw meat from the fridge, cut it, and skewer it, he just said, “Nope.”
In embodied AI, you are either bottlenecked by a technology that doesn’t exist yet, or you’re ignoring the basic user experience that has existed since 2010.
These two cases perfectly capture the paradox of the embodied AI PM. You are hired to define futuristic demand and drive product-market fit. Yet, you are fundamentally bottlenecked by a hard technical ceiling that makes even basic physical tasks impossible. This forces you into a corner: either you over-promise on robotics capabilities that will take a decade to mature, or you regress to fixing basic software-level UX flaws that robotics companies completely ignore.
Most people think the highest value of an embodied AI PM is defining cutting-edge robotic capabilities. That is a lie. The actual highest immediate value of a PM in this industry is aggressively de-scoping robot tasks and retro-fitting basic internet-era UX to physical machines.
Look at the delivery robot that lost my lunch. It lacked a queue visualization app. It lacked a push notification. It lacked a two-way confirmation system. These aren’t AI problems. They aren’t edge computing problems. They are Tuesday-morning-junior-developer problems. Any PM who actually walked the user journey would have solved this in a week. But no one did.
The robotics industry is obsessed with reaching the moon, but they haven’t figured out how to make the launchpad user-friendly.
The hard truth is that current embodied AI lacks the real-world physical interaction data required for generalization. We have a fraction of the data we need, and edge computing can’t run these massive models in real-time. Because of this hard technical ceiling, a PM cannot will a complex, long-chain task into existence. If you want a robot to autonomously make a BBQ skewer from scratch, you are out of luck.
But that doesn’t mean you do nothing. If you are a PM in this space, your job is to find the safety margin of current technology. You stop trying to define a robot that can do everything. You break the impossible task down into technically viable single points. Maybe the robot can’t skewer the meat. But can it move the meat from the fridge to the prep table? Yes? Great. Define the success rate, efficiency, and cost for that single point. Prove it works. Move on to the next single point.
Stop promising the autonomous future. Start decomposing the impossible into technically viable single points.
If you’re a PM in embodied AI, stop waiting for the tech breakthroughs. Stop acting like a sci-fi visionary and start acting like a plumber. Walk the user journey. Fix the doorbell. Build the app that tells me my food is waiting outside. Decompose the complex tasks into things your tech can actually handle today.
Your job isn’t to dream about the year 2035. Your job is to make sure the robot doesn’t steal my lunch today.
FAQ
Q: Isn't it the engineer's job to figure out the technical ceiling? Why is this the PM's problem?
A: Because engineers build what you spec. If you spec a fully autonomous robot that can't physically be built for 10 years, your product fails. A PM's job is to understand the technical boundary and carve out a viable product within it, not to wait for a miracle.
Q: What's the practical implication for a robotics startup today?
A: Stop trying to build end-to-end autonomous solutions. Find the narrowest, most painful bottleneck in a physical task that current tech can actually solve, and dominate that single point. Then, wrap it in a UX layer that actually communicates with the user.
Q: Are you saying embodied AI is a dead end?
A: No, the tech is just in its infancy. But the current PM approach is a dead end. If you have a robot that can navigate a building but can't send a push notification, you don't have a tech problem—you have a product management failure.