I have a small confession to make: I’m a bit of a nerd.
So, for now, something different from my usual posts: building a robotarm with AI.
About a year ago, I bought a 3D printer and thought it would be fun to create a robot arm.
I already had some experience with 3D design (I taught myself using an AutoCAD book when I was a kid). The tools have evolved (now Fusion 360), but the principles remain the same.
AI turned out to be a great co-creator. Whether it’s stepper motors, motor drivers, serial interfaces, or I2C magnetic encoders, it knows it all. It makes building much more enjoyable: I can focus on the design and the outcome, while AI helps with the details (and yes, being an architect helps 😉).
Together with GitHub Copilot, I was able to develop the software quite smoothly. During this project, a colleague once argued that (at that time, end of 2025) AI could only work within a single file, not across a larger codebase. From hands-on experience, I could show that this is no longer true (It is always good to come with data to an opinion fight 😉) With a well-designed spec, AI can work across multiple files and even across languages C++ (ESP32), Python (backend), and JavaScript/HTML/CSS.
Even in a one-person project, working spec-driven makes a big difference. It keeps the AI aligned and helps avoid drifting into the wrong direction.
That said: hardware is hard.
If a single cheap component doesn’t behave as expected, finding the root cause can be frustrating. AI helps, but without critical thinking and problem-solving, this project would not have succeeded.
Another lesson: physical reality matters. 3D printing is powerful, but tolerances (especially in gears and connections) are not perfect. You can compensate in software, but only up to a point.
I had to accept that my original idea, building a robot to play checkers, was not feasible with the chosen materials and components. The robot simply couldn’t pick up pieces reliably enough. So I adapted the scope and redesigned it to play Tic-Tac-Toe instead, which works (
https://lnkd.in/dVzU26Rr)
What I took away from this project:
- Spec-driven development works, even for smaller projects
- AI is a powerful co-creator, but not a substitute for critical thinking
- Hardware introduces constraints you cannot “prompt away”
- Build with change in mind: loosely coupled with clear API's (so changing from checkers to Tic-tac-toe was easy)
- Most importantly: being realistic about scope is key to actually delivering something
- Experimenting is great: it enables to come with data to an opinion fight
As architects, we often work on complex systems and ambitious ideas. This project was a small but practical reminder: focus on solving the real problem within realistic constraints and adjust when needed.