People ask me if I’m scared of humanoid robots. I tell them I’m scared of a world without them.
For the past decade, I’ve been obsessed with a single question: how can we build tools that empower people to do their best work? It’s what drove me to build RemoteTeam, which was later acquired by Gusto. We wanted to make it possible for anyone, anywhere, to be part of a great team. But what if the best team member for a certain job isn’t a person at all?
That’s the thought that kept me up at night. It’s the thought that led me and a small, crazy team to embark on a journey to build our first humanoid robot.
The Napkin Sketch That Started It All
It didn’t start in a fancy lab. It started on a napkin at a coffee shop in Hayes Valley. I was talking to a friend who runs a large e-commerce fulfillment center. He was telling me about the constant struggle to find people for physically demanding, repetitive tasks. The high turnover, the injuries, the sheer cost of it all. His best people were quitting after a few months, burned out from lifting boxes for 8 hours a day.
I’d seen it before. In my angel investing, I’ve backed over 200 companies, including some in the AI and robotics space like Scale AI and warehouse robotics companies. I’ve seen the power of AI to automate digital tasks. But the physical world? That’s a different beast. You can’t just write a script to move a box from a shelf to a conveyor belt. It requires a level of dexterity and adaptability that most machines just don’t have.
So I sketched it out. A robot that could walk, lift, and carry. A robot that could work alongside people, doing the jobs that people don’t want to do. It wasn’t about replacing people. It was about elevating them. I envisioned a future where the fulfillment center was a collaboration between humans and robots. The robots would do the heavy lifting, and the humans would do the problem-solving and the quality control. It was a vision of a more efficient, and more humane, workplace.
From Bits to Bones: Building the Body
We weren’t Tesla. We didn’t have a billion-dollar budget. We had a small seed round and a lot of grit. We started with off-the-shelf components. We bought motors from one supplier, sensors from another. We 3D printed parts in our garage. It was a Frankenstein’s monster of a robot.
Our first prototype couldn’t even stand up. It would just twitch and fall over. We called it “Wobbles.” It was a humbling experience. But with every failure, we learned something new. We learned about weight distribution, about the limitations of certain motors, about the importance of a low center of gravity. We spent weeks just trying to get the damn thing to balance. We tried different leg designs, different foot sizes, different control algorithms. Nothing worked.
We went through three major design revisions before we had a robot that could reliably walk on two legs. We used a combination of carbon fiber and aluminum to keep it lightweight but strong. We designed a custom gearbox to give it the torque it needed to lift heavy objects. We even developed our own battery pack to give it the power it needed to run for a full shift. It was a slow, painful process. But we were making progress. Each new version was a little bit better than the last. It was a powerful lesson in the power of iteration.
The Ghost in the Machine: Giving it a Brain
Building the body was hard. Giving it a brain was harder.
We knew from the start that we wanted to use a large language model as the core of our robot’s intelligence. I’ve been fortunate enough to invest in some of the leading AI companies like Anthropic, OpenAI, and Hugging Face. I’ve seen firsthand the power of these models. But using them to control a physical robot? That was a whole new level of complexity.
We started with a pre-trained model and then fine-tuned it on a massive dataset of human motion. We fed it videos of people walking, lifting, and carrying. We used reinforcement learning to teach it how to balance and how to interact with its environment. It was a massive undertaking. We had a team of five engineers working around the clock for six months.
There were times when I thought we were going to fail. The robot would do things that were completely unexpected. It would walk into walls. It would drop things. It would just… stop. One time, we were testing a new grasping algorithm. The robot was supposed to pick up a box and place it on a shelf. Instead, it picked up the box, and then threw it across the room. We never figured out why it did that. We just called it the “rage quit” incident and moved on.
The Simulation is the Key
We quickly realized that we couldn’t test everything in the real world. It was too slow, too expensive, and too dangerous. So we built a simulation. A virtual world where we could test our algorithms without breaking anything. We could run thousands of tests in a single day. We could simulate different environments, different objects, different lighting conditions. The simulation became our secret weapon.
It was in the simulation that we had our “aha!” moment. We were trying to teach the robot how to walk up a flight of stairs. It kept falling. We tried everything we could think of. Nothing worked. Then, one of our engineers had a crazy idea. What if we didn’t teach it how to walk up stairs at all? What if we just gave it a goal – to get to the top of the stairs – and let it figure out how to do it on its own? It was a long shot, but we decided to try it. We set up the simulation and let it run overnight. The next morning, we came in to find that the robot had learned to crawl up the stairs. It wasn’t pretty, but it worked. It was a breakthrough. We had unlocked the power of emergent behavior.
The First Steps
I’ll never forget the day it first walked. We had been working on a new control algorithm for weeks. We uploaded the code to the robot and held our breath. It stood up, took a step, and then another. It walked the length of the lab without falling. We all cheered. It was a moment of pure joy.
But it was also a moment of terror. We had created something that could move and think on its own. It was no longer just a machine. It was something more. It was the beginning of a new era.
The Future is Now
We’re still a long way from a world where humanoid robots are commonplace. But I believe we’re on the cusp of a revolution. These robots have the potential to transform every industry, from manufacturing and logistics to healthcare and elder care.
They’re not going to take our jobs. They’re going to take the jobs we don’t want. They’re going to free us up to do the things that only humans can do: to be creative, to be compassionate, to be human.
I’m not scared of a world with humanoid robots. I’m excited about it. And I’m proud to be a small part of making it a reality. The journey from that napkin sketch to a walking, thinking robot has been the most challenging and rewarding experience of my life. And we’re just getting started.
Frequently Asked Questions
How has this view evolved over time?
My thinking on most topics has changed significantly over the years. Early in my career, I held many conventional views that experience proved wrong. I try to update my beliefs when the evidence changes.
What experience informs this perspective?
This perspective comes from over a decade of building companies in Silicon Valley, two successful exits (RemoteTeam to Gusto, MovieLaLa to Gfycat), and investing in 200+ startups including Anthropic, OpenAI, and Scale AI. I write about what I've lived.
Do all experts agree with this view?
No, and that's fine. The best ideas in business are often contrarian. I share my perspective based on my experience and data, but I encourage you to seek out opposing viewpoints and form your own conclusions.