Today’s adventure with AI was an interesting mix of “wow, that’s helpful” and “hmm… maybe not that.”
I was fixing a small bug when I noticed the AI-generated saveHouses function was also calling setHouses() to set the state. The code still worked, but that hidden side effect isn’t ideal — it makes the logic less predictable and cause re-renders.
But at the same time, AI gave me a clean code review: it spotted redundant lines instantly, suggested a better edge-case handling, and even drafted a full architecture document with diagrams. Honestly, that part felt like magic.
✅ What AI Was Helpful For Today
- Pointed out redundant lines I missed
- Suggested a cleaner approach for an edge case
- Generated an architecture document + diagrams
⚠️ What AI Was Not Helpful For
- Added an unexpected side effect (setHouses() inside a save function)
- Needed human oversight to maintain clean architecture
Today reminded me that pairing with AI is a bit like pairing with a very fast junior developer: super helpful for cleanup and documentation, but still needs guidance on architectural decisions.
Top comments (1)
The diary format is a great approach for capturing evolving AI coding patterns. Your observation about knowing when NOT to use auto-mode is the most important insight here. I've come to think of it like pairing with a very fast but sometimes overconfident colleague: great for first drafts and boilerplate, risky when the problem requires nuanced judgment about your specific codebase. What's helped me: keeping a project context file that explicitly documents the places where auto-mode tends to get things wrong in our codebase. Pattern by pattern, the list helps me know when to stay in manual mode. Looking forward to diary entry #2.