Robots That Build Robots: A Conversation About the Exponential Future

The Recursive Loop

If you've used Claude Code, Cursor, or any agentic coding tool in the last six months, you already understand the concept. You give the model context. It builds something. You review. It iterates. The loop gets tighter and faster with every cycle. The tool literally gets better at building itself.

Claudio and I were on a call -- just a regular check-in about QWave business -- when the parallel hit me mid-sentence. The exact same recursive loop we use in software maps directly to robotics. Take context. Feed it to an LLM. Build a solution. Let the solution inform the next build. Recurse.

Robots All the Way Down

Here's where the conversation got wild. I started thinking out loud about a home care robot -- something that helps elderly people with daily tasks. Simple enough goal. But to build it, you need hydraulic components, precision springs, sensors, actuators. Each of those needs custom manufacturing equipment.

So AI designs the robot. Then AI designs the machines that fabricate the robot's components. Then AI designs the machines that build those machines. Robots that create the components necessary to create the robots, that create the robots. Es como una matrioska de automatizacion -- it's like a matryoshka of automation.

The progression is clear if you zoom out: simple task-specific robots (Roomba, welding arms) evolve into self-driving vehicles, which evolve into general-purpose humanoids. But the real inflection point isn't a better end product. It's when the manufacturing chain itself becomes recursive. That is the step change.

The Simulation Advantage

Here's the thing people don't fully grasp yet: physical testing is slow, expensive, and limited. You build a prototype, test it, measure the failure, redesign, build again. One cycle might take weeks. You're constrained by materials, lab time, physical reality.

But Google built 3D simulated environments with real-world physics over two years ago. You can model gravity, friction, material stress, thermal expansion -- every variable that matters. And in that simulated world, you can run a million tests on a robotic component in the time it takes to run one test in the real world.

You Don't Need to Be an Expert

This is the part that makes traditional engineers uncomfortable, and I get it. But hear me out. We don't have to know mechanical engineering, materials science, or construction methodology from the ground up. The knowledge already exists. It's in textbooks, research papers, patents, and the heads of experts who've spent decades in their fields.

Our job -- nuestro trabajo -- is to capture that expertise, distill it, and feed it into models that can apply it at scale. You take experts, you get them talking, you capture whatever context exists for that industry, you distill it, put it into a model, and then give it problems.

The Exponential Timeline

People ask me “when?” and I say ten years for full robotics saturation. But they hear “ten years” and think linear. They picture steady, gradual progress. That's not how this works.

Think about what happened in AI. From December 2024 to January 2025, when Opus 4.5 came out, the world changed. One model release. The same kind of inflection is coming for robotics -- we are just not in that industry yet, so we don't feel it the way we feel it in software. But it's coming. La curva es exponencial, no lineal -- the curve is exponential, not linear.

Position Your Business

Look, I am not saying you need to start building robots tomorrow. But the wave is coming, and the businesses that position themselves now will be the ones that ride it instead of getting swept under.

Here is your starting checklist. Every item here is something you can act on this quarter. Check them off as you go: