An engineer tested whether agentic AI could design a working mixed-signal circuit by asking Claude Opus 5.5 to create an MCU-controlled analog synthesizer from a manufacturer’s limited parts list. Starting with a broad concept and a written specification, the system designed a Juno-inspired digitally controlled sawtooth oscillator, a Moog-style transistor-ladder low-pass filter, and a voltage-controlled amplifier. It also produced circuit simulations, a behavioral audio model, schematics, firmware, and manufacturing files. The engineer reviewed the design but left its circuits unchanged, emphasizing that giving the AI a complete specification and moving through abstraction levels systematically worked better than repeatedly editing details.
The design took about 19 hours to reach an order-ready state, with the main bottleneck being PCB layout: two-layer routing failed, and a four-layer board required a functional-block floorplan and several rounds of human guidance. After fabrication, automated measurements showed that the oscillator, filter, and amplifier closely matched their SPICE simulations, and the audio loopback enabled calibration. The only hardware correction was removal of a solder bridge; touch controls needed recalibration and proved awkward to use. The result demonstrates that AI can produce a complex analog design that works in real hardware, while also showing that layout and practical usability still benefit from human oversight.
Summary generated by AI from the linked article. hn.today is not affiliated with Hacker News or Y Combinator.