AI vs. Human PCB Design Project Background

Circuit Mind is an AI engine built specifically for electronics. You enter your system requirements, build the architecture in a block diagram, set your constraints, and the platform outputs schematic sheets populated with components it selects. One of its main features is the integration with the Nexar API from Altium, allowing it to see real-time component availability across hundreds of distributors. This means it can suggest parts that are functionally appropriate as well as actually in stock.

This baseline design came from engineer Soheil Shabafrouz, who created an ESP32-PICO-based IMU PCB. The original design included power regulation, four separate sensors, wireless connectivity, storage, and a few extras like solder bridges and test points. In the video below, we tested the capability of Circuit Mind to replicate the schematic design created by Soheil.

Watch the original 1-minute design review on Altium Academy:

Power Supply

The two designs diverged in how they handled the main regulator.

• Human design: Stepped 4.2 V down to 3.3 V using a discrete buck regulator, an external inductor, and several passives—more complex, but highly configurable.
• AI design: Took a 5 V input down to 3.3 V using a highly integrated power module with a built-in inductor—simpler and more compact, but potentially pricier.

This is a perfect example of Circuit Mind's part-selection advantage. The integrated module is efficient in layout and assembly, but it’s easy for a human designer to miss unless they’ve seen it before. On the other hand, the discrete design might offer better cost control, performance tuning, or design for test capability.

Sensors

While the Circuit Mind's integrated sensor saves board space, it can also limit flexibility if one function underperforms or becomes unavailable. Another difference was in power delivery: the human version isolated the sensor rail with ferrite beads and filtering, while the AI tied all sensors directly to the main 3.3 V rail—arguably the safer default unless noise issues have been confirmed.

• Human design: Four distinct devices—accelerometer, gyroscope, magnetometer, and pressure sensor—each selected independently.
• AI design: Combined the accelerometer and gyroscope into a single chip, reducing component count to three.

ESP32 Microcontroller

Both designs implemented the ESP32 core functions and interface assignments correctly, but the AI added a few noteworthy elements:

• An impedance matching network between the antenna and LNA, with placeholder capacitor values for later tuning.
• ESD protection on the SD card lines to guard against insertion-related transients.

The human schematic, meanwhile, skipped those features but included a capacitor on the enable pin to shape startup timing—something the AI left out.

What the AI Skipped

Despite producing a complete and functional schematic, the AI omitted a few human touches:

• Solder bridges for reconfiguration and test access.
• Custom analog circuits tied to headers—components that exist outside of standard IC datasheets.
• LED indicators for user feedback.
• Embedded layout notes, parameter sets, and in-schematic design rules.

These omissions don’t make the AI wrong; they simply highlight that without explicit instructions, it won’t invent extra features that weren’t part of the functional requirements.

The table below provides a side-by-side comparison of both designs

The Verdict: AI and Human PCB Designers Should Work Together

There was no clear "winner" in this exercise. The human schematic was richer in optional features and field-ready flexibility, while the AI schematic leaned on modern integrated components and built-in supply chain intelligence. The best approach might be to treat the AI as providing a starting point, giving the engineer a path to a usable schematic. The human’s job is to refine the initial, add the extras, and prepare it for layout.

Instead of thinking about AI-based EDA software as a competitor, we consider it a collaborator. It accelerates the early design phase, suggests components you may not have known about, and ensures your BOM won’t be crippled by availability issues. Then, you step in to do what AI can’t: adapt the design for the real world, anticipate failure modes, and add the touches that improve usability and testability.