@nixCraft Is this really true at other universities?
At Aalborg University (department of #ComputerScience) we have allowed our students to use #GenAI in their projects since 2023. The project work is half of all the effort in the educations. For the courses it is mostly prohibited in the exams, but for the projects they just have to declare what how they have used it. They then defend the project at an oral exam.
We feel that this realistically mirrors the use in companies.
@nixCraft Now we have just realised that we need to change how we teach programming itself. So some form of programming using AI agents will be part of a course in the future.
On of the big questions is: How to use or not use AI in the very first programming course?
Please let me know if you do this, so I know not to bother interviewing any of your graduates.
@david_chisnall @nixCraft Are candidates will also learn to program by hand.
What are hearing from our local software industry is that they are utilizing AI tools as part of the programming workflow. This is the workplace reality that our graduates are coming to.
The question is. How do we bring our intake of students that consists of two populations. Those that have never programmed and those that dabble in vibe coding to a level where the can both code classically and use AI agents?
Perhaps it’s different for us because our customers care about security (and, by extension, correctness), but we do not allow LLM-generated code in our code bases at all. ‘Agentic’ flows lead to bad design choices that result in the need to churn code later, which is unacceptable if you want customers to build on top of your APIs. Any use of LLMs tends to produce code that looks correct but is subtly wrong in ways that code review can’t catch. Code review has been the bottleneck for my entire professional career and so a tool that makes it easier to generate code fast and makes that code harder to review is an absolute productivity killer.
That said, I am happy to enthusiastically cheer on our competitors adopting LLM-based tooling!
To add a little bit: universities should not be in the business of teaching tools. They should be in the business of teaching understanding that enables graduates to learn tools easily. A person who understands the way LLMs generate tokens and understands how to do software engineering can pick up a workflow that does or does not contain LLMs easily.
The same applies with programming languages. I don’t care what languages you used at university, I understand that you learned a diverse enough set that you covered the core concepts that you’re likely to encounter elsewhere and can pick up a new language quickly. This changes when I’m hiring senior people (where some domain expertise and experience is also expected) but fresh graduates are expected to know nothing and be able to learn anything. Anything they know is likely to be an oversimplification, or is a bonus in the rare cases where it’s actually useful.
When you incorporate tools that cause measurable harm in skill-acquisition ability into your curriculum, you are destroying the value that I want to see from fresh graduates.
"When you incorporate tools that cause measurable harm in skill-acquisition ability into your curriculum, you are destroying the value that I want to see from fresh graduates."
The students are already harming their own learning at the moment by using these tool during the course exercises and then failing the exam. That is the current state of affairs. We want to figure out how to teach them both to code with and without AI. And motivate them for learning both.
These days I am so very grateful for a couple things.
1. Retiring from the US Army well before any Trump administration (Obama was President when I left)
2. Retiring from software development well before the advent of AI
@nixCraft If my team lead approaches me today that I *must* use AI starting tomorrow, he can be damn sure I blow through the token limit by the time I finish my morning coffee.
He will get compliance, and life to regret it.
That’s depressing. Even if you accept the idea that using LLMs makes you a more productive programmer, the goal of a university is education, not productivity. As the analogy goes: you wouldn’t use a forklift truck to lift weights for you in the gym. The people who do seem to be getting some productivity boost from LLMs are experienced programmers who are able to very carefully shepherd them. Avoiding accumulating the skills that you get from constant practice is going to make you a less good software engineer with or without LLMs.
At school for me it was "no wikipedia!!". In 2026, it's one of the last good source online xD