Some thoughts on AI

AI is here to stay, that’s for sure. It’s a great tool for reviewing and refactoring code, writing tests, debugging problems etc.

But we should think of it as just that – a tool. Software development is a craft, and AI is one of the many tools in our belt. As with any tool, we shouldn’t it use it for everything.

Many junior developers are already so used to prompt based coding that you sometimes wonder how they would do without the constant help of AI. As AI gets better, the next generation of developers will probably never have the terrific experience of debugging a hard problem for days on end. They won’t learn from it as well. AI will raise to be a magic box that solves all the hard stuff.

But what if the AI writes buggy code? What if there’s a problem that’s so distributed in a system that the AI can’t connect the pieces? Most companies have at least one person with a very deep technical understanding of the project. A go-to person for the hard stuff. She/he often has a solid mental model of the project and knows exactly what code is there and how it works.

You won’t become this person by using AI for everything. Au contraire – you may not even have a clue what part of code a bug might be in, because you just write prompts and click “Accept”.

This isn’t a rant against AI – I use it myself every day. But please think of it as an assistant that you can’t really trust. Review everything it does. If you don’t understand why it wrote a certain line of code, ask it to explain it. Make sure that you have a solid understanding of the code.

My personal AI rules

If you can’t verify it, don’t use it.
Review each line of code, consider whether it is really the best solution.
Don’t trust it.