Working with AI
AI assistants are now part of everyday firmware work. At its core, this is about using them well: giving board and constraint context, being explicit about language and peripherals, and refining a prompt instead of accepting the first answer. Just as important is judging what comes back — invented registers, unsafe wiring, or outdated SDK calls — and never pasting secrets or unpublished schematics into a tool you do not control.
Starting Points
- Pereira, S. (2025). Generative AI for Software Development. O'Reilly.
- Official documentation and responsible-use guidelines for the AI tools you use (e.g. GitHub Copilot, Claude, ChatGPT).
- Working-with-AI fundamentals
- Responsible Use of AI
Key Points
- You use AI tools deliberately for common tasks (explaining a driver, drafting a state machine, suggesting tests, summarising a datasheet section you then verify).
- You write clear, structured prompts with board, language, and constraints, and you refine them iteratively.
- You critically evaluate AI output for correctness, electrical safety, and plausibility, and you recognise invented APIs or pinouts.
- You never share secrets, private keys, or confidential client data with an AI tool, and you document when and how AI was used.
- You treat AI-generated firmware as untrusted until it builds, runs, and is reviewed on hardware.