BAYOOMED has tested Github Copilot
Can artificial intelligence (AI) assist in software development?
Can artificial intelligence (AI) assist in software development?
Hardly any technology has produced so many pathbreaking innovations in recent months – we’re talking about artificial intelligence (AI).
One example from the field of medical software development is GitHub Copilot. As an AI tool, it supports software engineers in their coding. Copilot learns from the public code repositories on GitHub and can complete code, make suggestions and generate entire boilerplates for example.
Here at BAYOOMED, we wanted to find out: Do AI tools like GitHub Copilot deliver what they promise? Can we implement software projects more efficiently with Copilot and what are the weaknesses of the AI tool?
For our test, two teams worked on different projects during a one-month test phase. Team A, which focused on native mobile development using Android/Kotlin and iOS/Swift with new UI frameworks, found that Copilot’s suggestions were often unhelpful and not aligned with the frameworks’ usage. The tooling for Copilot in XCode was also less advanced.
In contrast, Team B’s assessment was positive. They were working on a cross-platform app using C#/.NET framework. After a brief learning period, GitHub Copilot suggestions accelerated their development, especially for repetitive tasks like writing tests.
CTO at BAYOOMED
Our test run with Copilot was accompanied by BAYOOMED CTO and AI expert Bernd Seidenspinner. Bernd has been part of BAYOOMED for more than 15 years and is a firm contact person for our team for all questions regarding new technologies in medical software development.
In short, we can confirm that GitHub Copilot can significantly improve the software development workflow, but its effectiveness depends on the workload. Copilot works best when the AI tool is used for coding familiar tasks. In contrast, GitHub Copilot’s weakness is clearly in working with unfamiliar frameworks.
And we couldn’t do it without our experts at BAYOOMED. GitHub Copilot does not require tool validation, because our software engineers review and validate code suggestions and output from Copilot to ensure accuracy, quality, and security before incorporating the code into their project. GitHub Copilot is not capable of autonomously generating substantial parts of software without the involvement of a software engineer to assemble and integrate the code.