.png)
2 hours ago
1
I am currently looking for a job! If you're hiring a new grad in 2026 for for Rust, TypeScript, or React, feel free to shoot me an email at serena (at) quamserena.com.
I don't use AI to code and I disagree with the notion that it can enhance developer productivity. I have found it extremely annoying — every time a suggestion pops up, I have to think about it to check if it's correct. Here's usually how it goes:
- Think about what I want to do
- Decide the way that I want to do it
- Start writing the code to do it
- AI suggestion pops up
- Pause the solution that I'm working through in my head to evaluate the new AI solution
- AI solution is different but looks correct
- Think about it more closely and realize that the AI solution is subtly wrong, missing some crucial insight, or just misunderstands the problem that I'm solving
- Have to context switch my brain back to my original solution
I don't understand the point. I switched AI suggestions off after getting fed up. I turn it back on occasionally, if only to verify that nothing's changed (indeed, nothing has), then turn it back off. AI is not a regular part of my workflow for writing code, and I think that it slows me down and doesn't benefit me at all. Maybe there is some secret technique to using AI to write good code, but based on the code that my classmates submit I don't think I'm missing much.
Sometimes prompting AI directly (not using suggestions) can be more useful. Google is keyword-based; LLMs have a much better understanding of the question that you're asking them. If you're working in a problem domain that you're not familiar with and you give ChatGPT a block of code (or math) and an error, it will generate a solution that implements some Concept that you didn't know about or misunderstood. The actual solution that ChatGPT generates is immaterial; the proper move here is to open up the documentation and start reading about Concept and implement it yourself. I have used this for learning OCaml and it's much nicer than trying to figure out the errors all alone. For stuff I already know, however, I don't use AI at all, and in neither case does AI actually write the code that goes into the project.
Most people don't use AI this way though. They (speaking of soon-to-be CS grads) use it as an excuse to avoid understanding the tools that they're working with. The result is that they write themselves into a corner, get some bug that they don't understand, and then pass it along to me. These bugs aren't that difficult and only take a few minutes to diagnose and fix. The horror here is that no one seems to understand what they're writing and simply don't have the skills to debug it when things go wrong. I'm fine with AI tools being used to boost productivity or whatever (if that works for you) but there is absolutely no way to avoid understanding the code you're writing.
