When people ask what I do, I say "AI/ML." When I look at my GitHub contributions, it's 60% TypeScript, 25% Python, and 15% config files. The reality of working in AI/ML as a student developer is far messier than the LinkedIn posts suggest.
The Decision Point
Third semester, I had to choose: double down on full-stack development where I was already comfortable, or pivot toward AI/ML where the ceiling felt higher but the floor felt nonexistent. I chose AI/ML, and I'd make the same choice again — but not for the reasons you'd expect.
It's Not About the Models
The reason I chose AI/ML isn't because transformer architectures are fascinating (they are) or because the job market is hot (it is). It's because AI/ML forces you to think about problems differently. In traditional SWE, you define the rules. In ML, you define the objective and let the data define the rules. That mental shift changed how I approach every engineering problem.
The Trade-offs Nobody Talks About
You need more math. Not "I took calculus" math. Linear algebra, probability, statistics, optimization — and you need to actually understand them, not just pass the exam. I spent an entire winter break re-learning probability from scratch because my intuitions were wrong.
Iteration cycles are longer. In web dev, you change code, refresh the browser, see the result. In ML, you change a hyperparameter, wait 45 minutes for training, and discover you made it worse. Patience becomes a technical skill.
The tooling is worse. React has incredible DevTools. PyTorch has print statements and hope. The ML ecosystem is improving fast, but it's still years behind web development in terms of developer experience.
Why I Still Write So Much JavaScript
Here's the dirty secret: most AI/ML work needs a frontend. Someone has to build the dashboard that displays model predictions. Someone has to create the API that serves inference results. Someone has to build the annotation tool that generates training data. That someone is usually me.
Being a full-stack developer who understands ML is more valuable than being a pure ML researcher or a pure frontend developer. The intersection is where the interesting work lives.
The Path Forward
I'm not abandoning software engineering. I'm augmenting it. Every system I build now has a question baked in: "Could this be learned instead of coded?" Sometimes the answer is yes, and I reach for scikit-learn. Most times the answer is no, and I reach for React. The skill is knowing which is which.
If you're a student deciding between SWE and AI/ML, my advice: don't choose. Learn both. Build systems that use ML where it makes sense and traditional engineering everywhere else. The future belongs to engineers who can do both — and the present has a serious shortage of them.