I remember the first time I saw a Boston Dynamics video. It was probably 2017, and Spot was trotting around, opening doors. My first thought wasn’t “Wow, cool robot.” It was, “We are so unprepared for this.”
Fast forward to today, and the conversation has exploded. We’ve got Figure 01 making coffee, Tesla’s Optimus folding laundry, and a dozen other startups promising to put a robot in every factory and home. As someone who’s been on the front lines of a few tech shifts—from the mobile boom with MovieLaLa to the remote work explosion with RemoteTeam—I can tell you this one feels different. Bigger.
I’ve been lucky enough to invest in some of the foundational companies in this space, like Scale AI and Anthropic, and I’ve had a ringside seat to the AI revolution. But building hardware is a different beast entirely. It’s not just about brilliant code; it’s about physics, supply chains, and a whole lot of things that can go wrong in the real world.
So, you want to build a humanoid robot? Buckle up. It’s one of the hardest things you can do in tech right now. But as an engineer and an investor, that’s exactly what makes it so exciting. This is the guide I wish I had when I first started exploring the space.
Don't Just Build a Robot, Solve a Problem
This is the number one mistake I see founders make. They get so obsessed with the technology—the actuators, the sensors, the walking gait—that they forget to ask the most important question: What problem are you actually solving?
A robot that can do a perfect backflip is an amazing technical demo. A robot that can reliably unload a dishwasher, 24/7, without breaking a single plate? That’s a billion-dollar company.
When we were building RemoteTeam (which Gusto later acquired), we didn’t start by trying to build the most complex HR platform imaginable. We started with a simple, painful problem: onboarding and paying international employees was a nightmare. We solved that first, then expanded.
The same applies to robotics. Don’t boil the ocean. Find a niche, a single, high-value task that is dangerous, dull, or dirty for humans to do. Some ideas to get you thinking:
- Logistics and Warehousing: Unloading trucks, sorting packages, last-mile delivery. Amazon is already a huge player here, but there are countless smaller warehouses that can’t afford a full Amazon-style automation overhaul.
- Manufacturing: Repetitive tasks on an assembly line, quality control inspection, moving heavy materials.
- Healthcare: Assisting nurses with patient mobility, delivering supplies in hospitals, or even surgical assistance (think surgical robots).
- Retail: Restocking shelves, cleaning floors, managing inventory.
Start with one specific use case and dominate it. Can your robot perform that one task better, cheaper, or more reliably than a human? If the answer is no, go back to the drawing board.
The Three Pillars: Hardware, Software, and Data
Building a humanoid robot is a multidisciplinary challenge. You need expertise in three core areas. You can’t succeed if one of these pillars is weak.
1. The Hardware: The Body
This is the most capital-intensive part of the equation. You’re dealing with custom parts, complex supply chains, and the unforgiving laws of physics. My advice? Don’t reinvent the wheel unless you absolutely have to.
- Actuators: This is the muscle of your robot. You have two main choices: electric or hydraulic. Electric actuators (like those in the Tesla Bot) are cleaner, quieter, and more precise. Hydraulic systems (like early Boston Dynamics models) are more powerful but also more complex and messy. For most commercial applications, electric is the way to go.
- Sensors: Your robot needs to see and feel the world. This means a suite of sensors: cameras for vision, LiDAR for depth perception, and force-torque sensors in the joints and hands to understand pressure.
- The End Effector (The Hand): This is arguably the most complex part of the robot. The human hand is a marvel of engineering. Replicating its dexterity is incredibly difficult. My advice is to start simple. A two-fingered gripper can handle a surprising number of tasks. Don’t try to build a five-fingered, human-like hand from day one unless your core use case absolutely requires it.
I’ve seen teams burn through millions of dollars trying to build their own custom actuators from scratch. Unless you have a truly revolutionary new design, it’s often better to partner with specialized suppliers for these components.
2. The Software: The Brain
This is where the magic happens. Your hardware is just a puppet; the software is the puppeteer. This is the domain of AI and machine learning.
- The AI Model: You’ll need a powerful AI model to control the robot. This is often a large language model (LLM) or a vision-language model (VLM) that can understand natural language commands and translate them into actions. Companies like Anthropic and OpenAI are building the foundational models that will power the next generation of robots.
- Simulation: You can’t test every new piece of code on a physical robot. It’s too slow and too dangerous. You need a robust simulation environment (like NVIDIA’s Isaac Sim) to train and test your AI models. You can run millions of scenarios in simulation in the time it would take to run a few dozen in the real world.
- The “Last Mile” Problem: Getting a robot to work 95% of the time in a lab is hard. Getting it to work 99.99% of the time in the chaotic, unpredictable real world is an order of magnitude harder. This is the “last mile” problem of robotics. It’s about handling edge cases: a dropped tool, an unexpected obstacle, a person walking in front of the robot. This is where you’ll spend most of your time.
3. The Data: The Fuel
Your AI model is only as good as the data you train it on. This is a lesson we learned over and over again at Scale AI. You need a massive, high-quality dataset of your robot performing its target task.
This creates a classic chicken-and-egg problem. You need data to train the robot, but you need a working robot to generate the data. How do you solve this?
- Simulation: Generate synthetic data in your simulator.
- Teleoperation: Have human operators control the robot remotely to perform the task. This is a great way to bootstrap your initial dataset.
- The Flywheel: As your robot gets better, it can start to perform the task autonomously. Every time it succeeds (or fails), it generates more data that you can use to retrain and improve the model. This creates a powerful data flywheel.
This is why I’m so bullish on companies that have a clear data acquisition strategy. The best hardware in the world won’t save you if you’re starving for data.
The Unspoken Truth: It’s Going to Be a Long, Hard Slog
I’ve painted a rosy picture of the opportunity, but I need to be brutally honest about the challenges. Building a humanoid robot company is not for the faint of heart.
- It’s Expensive: You’ll need tens of millions of dollars just to get a convincing prototype. The funding environment for hardware is much tougher than for software.
- The Timelines are Long: This isn’t a mobile app you can build in a few months. You’re looking at a 5-10 year journey to get to a commercial product at scale.
- The Talent is Scarce: The number of people in the world who have successfully built and shipped a robot is tiny. You’ll be competing with Google, Tesla, and a dozen other well-funded startups for the best talent.
But despite all of that, I’ve never been more optimistic. We are at the very beginning of a new industrial revolution. Just as the internet and the smartphone changed every aspect of our lives, humanoid robots will do the same for the physical world.
My final piece of advice is this: stay focused. The temptation to add more features, to make the robot more general-purpose, will be immense. Resist it. Solve one problem, solve it better than anyone else, and then, and only then, expand your ambitions. The journey is long, but the impact you can have is immeasurable. Now go build the future.
Frequently Asked Questions
How do I measure success with this approach?
Pick one or two metrics that directly tie to your goal and track them weekly. Vanity metrics like page views or follower counts rarely matter. Focus on metrics that reflect real engagement or revenue impact.
Do I need technical skills to build a humanoid robot (the guide i wish i had)?
Not necessarily. While technical understanding helps, the most important skills are clear thinking and the ability to break problems into smaller pieces. Many successful founders I've invested in started with zero technical background and either learned enough to be dangerous or found the right technical partner.
What tools do I need to get started?
Start with the basics. You don't need expensive software or fancy tools. A spreadsheet, a note-taking app, and direct access to your customers will get you further than any enterprise platform. Add tools only when you hit a specific bottleneck.
What are the most common mistakes when building a humanoid robot (the guide i wish i had)?
The biggest mistake I see is overcomplicating things early on. Start with the simplest version that works, get real feedback, and iterate from there. Another common trap is copying what worked for someone else without understanding the context behind their decisions.