My Friendenemy, Claude

As a developer in a startup company we are accustomed to deliver features rapidly. Last December, our entire sprint was dedicated to a massive overhaul, redeveloping our legacy admin interface from scratch. The codebase carried the history of every developer who had touched it over the years, making it a pretty complicated puzzle to solve. But surprisingly, we finished it in just a couple of months. It was spearheaded by three amazing developers, and I joined in as the reinforcement toward the end. Sitting there, watching how fast we were moving, I finally saw it for myself: Claude can create miracles. Claude is magic.

But as the sprint wrapped up, I didn't feel that usual win. I wasn't fulfilled. I started feeling this weird dilemma with "vibe coding" where you realize you don't actually own the code anymore. The "art" of development had shifted entirely into the testing phase; when you don't write the logic yourself, you have to spend so much more energy double checking everything just to be sure it's actually done. Even for a minor UI bug, my first response was to just jump into the terminal and write a prompt. It felt like it was actually slowing down my natural progress. With all that frustration building up, the only thought in my head was: I have less confidence because of AI. I actually blurted it out during our retrospective meeting and let the team discuss it openly. My main takeaway from that session was that we have to embrace this new landscape. AI is taking over the world, and the worst thing we can do is ignore it, especially since it speeds up development and prototyping so much. We just need to be mindful of how we use it. Still, those perspectives didn't sit right with me immediately. They didn't even come close to fixing the lack of confidence I was feeling while using AI. To me, it still felt better to be the one actually coding, because then you know the ins and outs of every feature. You can actually be proud of what you built.

Then, that same afternoon, everything changed. Due to some team restructuring, I was told I was moving to the backend. I couldn't help but be happy because I've wanted to do backend for the longest time, it felt like a dream come true. But when the actual development started, my lead gave me a high-level overview and then it was time for the real battle to begin. That's when I realized I had no choice but to truly rely on Claude. Suddenly, I found myself developing a new rhythm: I couldn't just "use" AI; I had to command it. I started by changing my entire workflow. Instead of just asking Claude for code, I realized I needed to give it the right context first. I became really disciplined about writing READMEs and technical markdown files before even starting a task, because that served as the "brain" for the AI to understand our complex system. I was so reliant on this process that I actually burned through my daily tokens every single day; there was even a time when I had to pay for extra just to keep my momentum going. It turned into a solid process: I'd start with a technical planning doc to map out the logic, then I'd implement and test everything phase by phase. I was churning out markdown files for every step of the development just to keep everything organized and documented. With my teammates constantly reviewing my work, I was able to deliver features a lot faster than I expected. It feels like I'm back in my junior years, with a Copilot by my side helping me get comfortable with the syntax until I'm ready to take over and write it myself. It hit me then that the way we develop has completely shifted. Before, you'd have to watch multiple tutorials and read through endless documentation, starting small before you could ever really contribute. Now, with the help of AI, you can finish tasks easily and just learn as you go.

I think my real problem was that I wasn't receptive to Claude when it came to frontend engineering. I've been in the industry for almost four years, and I couldn't accept that I wasn't doing things the "usual" way anymore. I'm knowledgeable in that aspect; I know I'm capable of solving those problems, and I find it fulfilling to design and solve a problem myself. But as a backend engineer, I don't know as much yet, so I'm okay with trusting Claude with my integrity as a developer. My end point is really just this: confidence in AI is contextual. The sweet spot for me is still that mix of AI and human intervention. You will only feel truly confident if you are also knowledgeable about the tool you are using. Moving forward, I need to find that balance in the frontend by using AI without losing my craft, while in the backend, I'll keep leaning on that balance of AI assistance and human guidance.

Well, I've said quite a lot already. Bye!