I spent two decades designing analog and mixed-signal integrated circuits. Then I became fascinated by the moment an AI decision turns into a real action. So I built AI_CUBE: two servos, a 3D-printed frame and a local brain. The closer software gets to hardware, the less room there is for uncertainty. The LLM chooses intent. Deterministic software controls the motion. Primitive → Capability → Agent Action → Guard → Executor. The next loop adds vision, audio and feedback from the physical world. AI_CUBE is not the destination. It is a test bench for agentic engineering. The same principles may one day guide local AI inside analog design flows. Give AI freedom where it helps — and hard boundaries where engineering begins.

I spent two decades designing analog and mixed-signal integrated circuits. Then I became fascinated by the moment an AI decision turns into a real action. So I built AI_CUBE: two servos, a 3D-printed frame and a local brain. The closer software gets to hardware, the less room there is for uncertainty. The LLM chooses intent. Deterministic software controls the motion. Primitive → Capability → Agent Action → Guard → Executor. The next loop adds vision, audio and feedback from the physical world. AI_CUBE is not the destination. It is a test bench for agentic engineering. The same principles may one day guide local AI inside analog design flows. Give AI freedom where it helps — and hard boundaries where engineering begins.