I used to work as a frontend developer, but I now work in QA automation. At work I use Claude heavily: I design the automation architecture, while AI writes much of the implementation. It has made me more productive, but I recently realized I haven't completed an entire ticket manually in more than six months.
That made me worry about technical interviews and whether I could still write code confidently from scratch. Coding is a skill that needs practice, so I started a small Vue application with a Playwright automation suite. I'm building it manually using documentation and search, without AI assistance. I'm treating it more like a hobby and a chance to practice than an efficiency project.
Do you do anything similar? If you use AI for most of your work, how do you keep your hands-on coding skills sharp?
4 Answers
I use AI mostly for paid work, but I leave it out of personal projects. I’ve built small things like a budget calculator, a socket-based chat room, and an inventory app without it. For me, the point of a side project is enjoying the process, not maximizing output. Using AI to write everything would feel a bit like hiring a chef when you enjoy cooking.
I use AI heavily for personal projects because otherwise many of my ideas would never get built. I treat it as a junior developer: I handle the architecture, requirements, and decisions, and I’m responsible for understanding and testing what it produces. For learning projects, though, I deliberately work manually on unfamiliar concepts so I don’t outsource the part I’m trying to practice.
That seems like a sensible split. AI is useful for exploring ideas and removing tedious work, but it’s worth keeping regular no-assistance practice—especially before an interview.
I don’t use AI for personal projects because the learning and the struggle are the main reasons I’m building them. I still use documentation, search, and examples, but I want to understand the mechanics rather than just accept a working answer. A small project built manually is also a good way to prepare for interviews.
I’ve started doing short practice sessions too: a blank project, no AI or autocomplete, and one small feature in about 45 minutes. It exposes gaps much faster than reviewing a large codebase that an agent helped create.
There’s a middle ground. I’ll build the first version and the important logic myself, then use AI for repetitive work such as test scaffolding, data imports, refactoring suggestions, or research. I still review every line and avoid using it for areas I’m actively trying to learn. Reviewing code helps, but it doesn’t replace actually writing code from scratch.
That distinction matters. Generated code can help you move faster, but writing the solution yourself exercises a different skill than inspecting or correcting someone else’s implementation.
That’s exactly the feeling I’m aiming for. The slow part is frustrating sometimes, but I remember the solutions much better when I work through them myself.