AI enters electronic product development through two doors. The first is the product itself: devices that detect, analyze and decide. The second is the development process: how code is written, tested and documented. At ProtoMakers, AI already runs inside a product we developed for a client, and it is a standard part of our development and testing process. This is what we learned.
1. AI inside the product (Edge AI)
Modern microcontrollers, such as the ESP32-S3 and newer STM32 families, can run small machine-learning models directly on the device, with no cloud. This opens possibilities such as:
- Fault detection: the device learns what normal operation looks like and alerts when something deviates.
- Predictive maintenance: analyzing vibration, current or temperature to detect wear before it becomes a failure.
- Voice and image recognition: voice commands or object detection in a camera, on the device itself.
When it doesn't fit: when a simple algorithm does the job. A moving average and a threshold solve many problems cheaply and reliably. That is why we always start with a feasibility check.
2. AI-accelerated development
AI coding tools have improved a lot. In firmware development they help mainly with:
- Writing routine code: drivers, protocol parsers and peripheral configuration.
- Reading long datasheets and finding the right detail.
- Code review: AI finds common bugs, such as a variable missing
volatileor a timing issue between interrupts. - Documentation that stays up to date with the code.
The result is less time on routine work and more time on the engineering decisions that really matter.
3. System testing with AI
HIL (Hardware-in-the-Loop) tests connect the real board to a setup that simulates its environment and run automated scenarios on it. AI adds analysis of large amounts of logs and measurements: detecting anomalies, comparing versions, and finding patterns a human eye misses in a file of a million lines.
Where AI doesn't replace an engineer yet
- Responsibility: someone has to understand every line of code running in a medical or defense device. That is still an engineer.
- The hardware: AI doesn't see the board. Electrical noise, a connector wired backwards or a part that runs hot are found with an oscilloscope and a multimeter.
- Design decisions: choosing the architecture, balancing cost and performance, and understanding what the client really needs.
This is also why so many people who planned their product with ChatGPT still need an engineer to build it. You planned with AI. What next?
Summary
AI doesn't replace the engineer, but an engineer who works with AI develops faster and tests more deeply. Tell us how AI could help your product.
Written by Roman Alperovich, founder and lead engineer at ProtoMakers.