The convergence: AI plus robots plus autonomy

I’ve been writing about these technologies separately for seven years. AI in one post. Robots in another. Self-driving in another. Chips in another. Like they’re different stories.

They’re not. They’re the same story.

The threads

AI gives machines intelligence. The ability to understand language, reason through problems, learn from examples, generate plans. Two years ago, this meant chatbots. Today it means agents that can browse the web, write code, book flights, and make decisions.

Robotics gives them bodies. Arms, legs, hands, sensors. The ability to exist in physical space, manipulate objects, navigate rooms, carry loads. Two years ago, this meant demo videos. Today it means factory deployments.

Self-driving gives them mobility. The ability to move through the world safely, read road conditions, anticipate other agents, make real-time decisions at 65 mph. Two years ago, this was a pilot program. Today it’s 100,000 rides per week.

Chips give them speed. The ability to process information fast enough for real-time decisions. TSMC at 2nm. NVIDIA inference chips. Apple custom silicon. The compute per watt per dollar is improving at a rate that makes yesterday’s impossible into tomorrow’s embedded.

The convergence

Here’s what happens when you combine them.

A Figure robot with a language model can understand a verbal instruction, plan the physical steps, and execute them. That’s AI plus robotics.

A Waymo car with an AI model can navigate construction zones it’s never seen before by reasoning about the scene in real time. That’s AI plus autonomy.

A robot that can navigate a factory, understand spoken requests, manipulate objects, and learn new tasks from video. That’s all four threads converging into a single machine.

And these machines are getting cheaper. The same manufacturing learning curves that made smartphones go from $1,000 luxury items to $200 ubiquitous tools are starting to apply to robots and autonomous vehicles.

What emerges

When the threads converge, capabilities emerge that none of the individual technologies possess. An AI model alone can’t pick up a box. A robot arm alone can’t decide which box to pick up. A self-driving algorithm alone can’t work in a warehouse. A chip alone does nothing.

Together: a machine that can understand what needs to be done (AI), do it (robotics), move to where it needs to be (autonomy), and do all of this fast enough to be useful (chips).

That’s not a tool. That’s a worker. A physical agent in the world that can think, move, decide, and act.

I’ve been careful about the word “worker” because it implies job displacement and that’s a conversation society isn’t ready for. But avoiding the word doesn’t change the reality. The convergence is producing machines that can do physical work that currently requires human workers.

The timeline

The pieces are on the table now. All of them. Anthropic and OpenAI supply the intelligence. Figure, Tesla, and a dozen others supply the body. Waymo, Tesla FSD, and Aurora supply the navigation. TSMC and NVIDIA supply the compute.

Assembly is underway. Not finished. Underway. The integration is happening in real time, in factories and on streets and in research labs.

Five years from now, I don’t think I’ll be writing about these technologies separately. I think they’ll be one thing. A class of machines that think, move, decide, and work. And the conversation won’t be about whether they exist. It’ll be about what their existence means for the rest of us.

I don’t have that answer yet. I’m not sure anyone does. But the question is no longer theoretical.

Related thinking: