Yesterday I realized I would need to send emails from a new app I'm building, and I remembered I had already integrated a previous service with Oracle Cloud to send email verifications. My conclusion was: let's create a microservice to send emails.
Yeah, I know that is not a very good decision and I could just integrate the email provider in both services. But I thought: this will look really good in my GitHub profile. In any case, that is not the point of this article.
I told Claude Code my idea and asked it to generate a full technical requirements document based on the already existing integration. Around 300 lines of specs later it seemed really well structured and documented, so I decided to start building it. 5 minutes later I didn't know how to do a specific thing, and I asked CC how I would do it. Instead of answering, it created the perfect implementation of the authentication system I was working on. That felt wrong. I was doing that project because I wanted to enjoy building it. Now a crucial part of the service had appeared out of thin air. In less than 5 seconds...
Some time later I gave up. I just asked Claude to follow the document from start to end and implement it. It took 15 minutes... and I felt honestly sad. I tried reviewing the implementation, trying to come up with something wrong to feel better about myself for knowing "more" than the AI. But I couldn't.
However, after walking the dog, I decided to completely ignore the AI implementation and do it myself. This was great! I started understanding the decisions the AI took, and in some cases I came up with better alternatives (at least ones that felt easier for me). After a while I started comparing my implementation with Claude's, and honestly, mine was so much better in a lot of architectural decisions. Each time I found a function I could do better, a shot of dopamine reached my brain. Another important point was that I had more knowledge of the rest of services, for example the rate limiting is done in my API gateway so there was no need to implement it here.
I didn't complete the project, not even close, but man, I enjoyed it. I'm excited to continue it tomorrow and learn more things.
Yes, AI is amazing. It can do a lot of things faster and sometimes better than us, but does it really matter if you don't enjoy it? I recommend anyone who likes coding and has been hooked on Claude Code, Codex, Cursor, or whatever AI tool, to spend an afternoon building something themselves.