I cut my AI spend and doomscrolling in one week (solo dev playbook)

Last month I had two ugly habits as a solo builder: I was burning tokens like crazy while vibe-coding. Every time I got stuck, I opened social feeds "for 2 minutes" and lost 45. If you’re shipping with Claude/Cursor/Codex every day, you probably know this combo. Here’s the exact playbook I used to fix both in 7 days. 1) I stopped flying blind on AI cost My biggest mistake was thinking I had a model problem. I didn’t. I had a visibility problem. I kept asking: “Why is my bill high?” “Was it this prompt or that prompt?” “Did I really use that many tokens?” Without live feedback, every session felt cheap… until invoice day. What I changed I started running TokenBar in my menu bar while coding. It’s a $5 Mac app that shows token/cost usage in real time across my AI sessions. That one tiny UI change did 3 things: I caught waste immediately (looped prompts, retry spam, giant context dumps). I switched to smaller prompts faster. I treated each run like a budgeted decision, not a casino spin.