Cool demo, but letting an AI Agent drive a physical probe may have problems. AI is probabilistic, but hardware is precise. If the model miscalculates a pin's position by even 0.1mm, the probe may crush the board. I am curious how the author actually bridges the gap between the Agent’s 'guess' and the sub-millimeter precision needed to avoid damaging hardware?