Does GitHub CoPilot Help Embedded Developers?
AI-powered coding assistants like GitHub Copilot and ChatGPT are transforming software development—but what do they mean for embedded systems engineers? Unlike traditional software, embedded programming comes with unique constraints such as limited resources, real-time requirements, and hardware dependencies.
This talk explores the real-world impact of GitHub Copilot on embedded development. We’ll separate hype from reality through practical examples, showcasing where AI tools can truly boost productivity—and where they fall short. Whether you're skeptical or excited about AI-driven coding, this session will help you understand how (and when) to make these tools work for you.
What is the primary advantage of GitHub Copilot's integration into VS Code that Reinhard highlights for embedded developers?
CoPilot learns from open source projects or public information.
AutoSAR and Vector tends to be closed source. Hence it will be weak.
Thanks for a very informative session.
In my experience I found github copilot to be very weak and ineffective in deeply embedded eco-systems like
- AUTOSAR + OSEK OS
- Vector toolchains + CAPL
so my takeaway is, any environment involving lots of standard documentation and having complex interconnection within them is a tough terrain for AI in its current state.
It might also have to something to do with closed source documentation (vector eco-system for example)
Thanks for the nice presentation! Would you please upload the slide so that we can easily access links, mentioned in the presentation.
The slides are now available for download
Excellent presentation Reinhard!
Thank you very much Reinhard. Our team recently used github copilot to build a firmware requirement document for a redesign of a 25 year old legacy product. It saved us months of time otherwise would have spend understanding what the old code is doing. Im still skeptical about using AI to generate code and (then developers reviewing it) though some organizations are pushing towards that direction. I feel like it can eventually degrade the quality of the human input in the process and offloading cognitive skills will pose more risks in the future like plateauing innovation. Any thought on that ?