Robotics does not have a brain problem. It has a parts problem. The popular story says one general foundation model is about to arrive, drop into a humanoid body, and turn every robot into a capable worker overnight. I review toolkits for a living, and that story skips the part where somebody has to build the toolkit first.
That is roughly the argument from Amulya Vishwanath, who spent seven years leading product marketing for NVIDIA’s robotics platforms before moving into venture capital. In a recent TechCrunch interview, she described robotics as being in its “Lego era.” The basic building blocks, including actuators, robotic hands, software, and safety standards, are still being developed. In her framing, the industry is still making the bricks, not snapping together finished sets.
Coming from someone who sat inside one of the companies most associated with the AI boom, that is a useful reality check. It also matches how I think about any new category of tools.
Why the ChatGPT comparison breaks down
ChatGPT felt sudden, but it landed on a foundation that already existed. People had keyboards, browsers, phones, and decades of digital text. The model was new. The interface was familiar. Nobody had to invent a new kind of finger to type a prompt.
Robots do not get that shortcut. A model that understands a task still needs a body that can carry it out, and that body depends on hardware that, by Vishwanath’s account, is not settled yet. A smart model attached to an unreliable hand is still an unreliable robot. Software can be patched overnight. A gripper that cannot handle a delicate object needs a redesign.
This is the gap between a demo and a product. Demos show what is possible under good conditions. Products have to work on a bad day, in a messy room, around people who did not read the manual.
What the “Lego era” looks like from a reviewer’s chair
When I evaluate a toolkit, I ask a few plain questions. Does it work out of the box? Do the parts fit together? Is there a standard way to connect it to other things? Can I trust it not to break something important? Applied to robotics today, Vishwanath’s list reads like a checklist of open items:
- Actuators: the motors and mechanisms that turn commands into motion are still being refined.
- Robotic hands: manipulation remains one of the hardest pieces to get right.
- Software: the layers tying perception, planning, and control together are still taking shape.
- Safety standards: the rules for putting robots near people are still being written.
None of that is a knock on the field. It is simply where a young category sits before it becomes boring and dependable. And boring and dependable is exactly what buyers want.
Safety standards deserve extra attention. A chatbot that gets something wrong produces a bad answer. A robot that gets something wrong produces a bad outcome in the physical world. That difference alone suggests robotics will move on a slower, more careful timeline than text-based AI, regardless of how good the models become.
The hype is not coming from nowhere
It is easy to see why people expect a breakthrough. Humanoids are showing up at industry events, including a Unitree Robotics G1 demonstrated at the Humanoids Summit in Tokyo in May 2026. NVIDIA CEO Jensen Huang opened CES in Las Vegas with a presentation covering the Rubin platform, open models, and autonomous driving. NVIDIA has also been talking publicly about a future where AI agents and robots use computers alongside people.
Put those images together and the narrative writes itself: big compute, big models, walking robots, and a breakthrough just around the corner. Vishwanath’s point is not that this future is fake. It is that the visible layer, the impressive demo and the big keynote, sits on top of an unfinished stack.
Why this matters for builders and buyers
If you are building in robotics, the “Lego era” framing is good news in disguise. When the bricks are not standardized, there is real room for teams who make better bricks. A more dependable actuator, a more capable hand, cleaner software tooling, or clearer safety practices could matter more right now than another general model announcement.
If you are buying, or just watching, the takeaway is to calibrate expectations. Treat polished humanoid demos the way I treat a slick product page. Interesting, sure. But I want to know what happens on day thirty, not day one.
It is also a reminder that “ChatGPT moment” may be the wrong yardstick. Software can go viral because distribution is nearly free. Hardware has to be manufactured, shipped, installed, maintained, and certified. Even a real leap in robot intelligence will likely spread at the speed of factories and safety reviews, not app downloads.
My take
I think Vishwanath has this right, and I appreciate hearing it from someone with seven years inside NVIDIA’s robotics work rather than from a skeptic on the sidelines. The foundation model story is exciting, but models are only one brick. Until actuators, hands, software, and safety standards mature, robotics will keep producing great demos and limited products.
That is not a reason to tune out. It is a reason to watch the unglamorous parts closely. When the bricks finally become standard and reliable, the sets will come together fast. My advice is to pay attention to the parts suppliers, not just the people on stage. That is usually where the real story starts.
đź•’ Published: