3 Ways I Use LLMs As a CS Student

I'm graduating from college soon and unless I go back for my master's at some point, this may very well be the last post I write as a student. In this post, I want to share three ways I've been using Claude Code as a student. It's in contrast to how I've been using it as a professional software engineer, which I'll probably write about in a future post.

1. Learning New Technologies

As a student, you end up learning a lot of new technologies, including programming languages, frameworks, and dev tools. However, classes rarely go that deep into teaching you how things work at a deeper level. Think about it: for any given class, you're in lectures for maybe 3 hours a week, and if you're anything like me, you're probably not retaining most of the information anyway. You're basically learning just enough to get by and then forgetting it right after the semester ends. This is especially true for classes that use a lot of different tools, like web development classes where you might be using React, Spring Boot, Docker, and a bunch of other stuff all in one semester.

Before LLMs, if I wanted to learn more about a technology, I would have to sift through documentation, Medium articles, and YouTube tutorials just to get my footing. While I still think documentation is the best way to learn (especially "Getting Started" guides), LLMs have made it so much easier to get answers to specific questions and to understand things more deeply.

For example, when I was first learning React, I didn't understand why calling setState(state + 1) twice in the same method would only increment the state by 1. I eventually came across this StackOverflow answer, but I didn't understand half the words in it. If I had an LLM at the time, I could've just asked it to explain the answer in terms I'd understand, and then follow up with more questions until I fully understood what was going on behind the scenes.

LLMs are such a good tool for learning how things work, and I think it's a shame that most students don't use them for this and just use them to generate code without understanding any of it. In my opinion, LLMs have made it so easy to learn a new stack that there's no longer any excuse to not understand the tools you use at (at least) a surface level. Stop copying and pasting code and calling it a day, use LLMs to help you understand it.

2. Building Simple Tools

With the rise of actually-good coding agents, the gap between "I wish I had a tool that could do X" and "I have a tool that can do X" has shrunk dramatically. For one of my classes, we are required to design a poster and present it at the end of the semester. One of the poster requirements was to include a section on testing, which I wanted to display in a radial chart format. After perusing for a while, all the tools I found for making radial charts were either pay-to-use or broken (as in the site would crash when I tried to change inputs). A few years ago, not having a tool like this would've been a huge blocker for me, but with Claude Code, I was able to generate a working radial chart generator in about 15 minutes of prompting.

Could I have built this myself? Probably, but definitely not in 15 minutes. For simple apps like this where the requirements are straightforward and the stakes are low, building with an agent as effectively closed the gap between an idea and a working prototype. Next time you need a quick tool and don't have it, give Claude a go.

You can check out more of these tools here.

3. Doing the Stuff I Already Know How To Do

I might catch some heat from teaching staff for this, but I think it's worth saying: if I have an assignment that I already know how to do, I'm probably going to use an LLM to do it. For example, a lot of web development classes have tedious work where you're just copying and pasting CRUD. Over and over. At this point in college, I could do that shit in my sleep. And with how good LLMs are getting, it's a part of the process that requires little to no human involvement anyway. I'll still review the code of course, especially if other people will be working on the same codebase later.

There are just better things to spend your time on. For example, designing the backbone of the CRUD operations (REST API endpoints, database schemas, etc.) is a much more enriching exercise. However, I must admit this is a slippery slope. Once you become okay using LLMs to generate some code for you, it becomes easy to let it generate more and more until it reaches a point beyond your understanding.

My advice here is to learn what your limits are and to keep pushing them. If your LLM-generated code is reaching a point that exceeds what you can read and understand, it might be time to go back to step 1 and learn a bit more. Become so knowledgable in an area that everything an LLM spits out is easy for you to understand and modify. Then you can use LLMs to do the heavy lifting while you focus on the high-level design and understanding of the code.