Your Moulinex is not replacing my kitchen staff
Hanging out in Discord servers with software engineering students, I sometimes hear of the adoption of LLMs, like Cursor or Copilot, by their teachers. The teachers are reported to be free from having to type code by hand, allowing them to focus on research and teaching. This bothers me a bit.
Students themselves are looking at this and come to believe that the time of the software engineer is up, since the people they regard as experts seem to be convinced that this is the future.
University teachers and researchers are experts at navigating the world of academia and focusing on research projects that reach somewhat of a conclusion after publishing. The publishing of theoretical papers and a high h-index are what most academics are judged on and so it is normal for them to optimise that part of their work.
However, this is not what is expected of a prospective software engineer. Development of and caring for software systems over several years require a deeper understanding of the technical details, the design and the business domain.
Obviously, a laughable amount of professional cooks (if any) have lost their jobs due to a multi-cooker replacing them. That is because there is a difference between the needs of a professional kitchen and home cooking.
In this metaphor, the code produced by academics is akin to the needs of home cooking: tailored for smaller needs and done by people who are overworked with doing a myriad of other stuff.
The multi-cooker, a piece of hardware priced at over a thousand euros, will definitely help you feed a household, but its usefulness cannot be easily scaled to meet the needs of a professional restaurant with four hundred seats that serves two thousand people a day.
Students who get told by their teachers that they should be using an LLM for coding must come to the realisation that they are not attending a culinary institute but a home-cooking class: teachers are using an LLM because the production of code is a minor step in achieving their goals (furthering the state of research), and for which they usually rely on dedicated research engineers.
In short:
- Researchers are not software engineers by trade, but are tasked with teaching students who take this path;
- The usage of LLMs as a convenience for teachers is ruining the understanding of students about what will be asked of them once in the workforce;
- The practice of software engineering – not merely prototyping to get approval by an auto-grading system – must be taught, not just seen as something for students that are not good enough for research.
ChatGPT cannot help you for shit in refactoring a ten year-old codebase. An industrial-grade compiler will. Understanding of the business domain by speaking with the experts will. A senior engineer will.