Two types of devs in 2026:
→ Debugs code with AI
→ Debugs AI’s code
Both are valid. Only one sleeps peacefully after deployment. 😅
I’ve been both this year.
Asking Codex “why is this API slow?” → quick answers, decent fixes. ✅
Reviewing Codex PRs after a vibe-coding session → “why are we fetching count + data in the same query again?” 🤔
The shift is real:
- Writing code is easier
- Reading code is harder
Because no one actually wrote it. It was generated → accepted → shipped. Then prod happens. 🔥
AI isn’t the problem. Blind acceptance of AI is.
The real skill now — can you read a diff and say: “I trust this at 10x traffic”? ⚡