Every night I told myself I had two separate problems:
- My AI coding bill was creeping up.
- My focus was getting wrecked by feeds.
Turns out they were the same bug.
When I was distracted, I wrote sloppy prompts.
Sloppy prompts created reruns.
Reruns burned tokens.
Then I’d panic, context-switch, and burn more.
What changed
I started tracking two numbers per session:
- token/cost burn for the current build block
- how many feed-triggered context switches I made
That single habit changed how I work more than any model switch.
The simple workflow I use now
1) 90-second preflight before I run anything
I write a tiny brief:
- what I need
- what files matter
- done condition
This alone cut expensive “one more try” loops.
2) Hard cap on reruns
If I’m on run #3, I stop and rewrite the brief.
No exceptions.
3) Feed-free build blocks
I don’t block the whole internet.
I just remove the algorithmic feeds that trigger scroll spirals.
4) End-of-session handoff note
I leave 4 lines for future-me so I don’t restart cold tomorrow and burn tokens reloading context.
Why I built tools for this
I made two tiny Mac apps because I wanted this behavior to be effortless:
- TokenBar ($5): live token/cost visibility in the menu bar → https://tokenbar.site
- Monk Mode ($15): feed-level distraction blocking on Mac → https://mac.monk-mode.lifestyle
Not magic. Just guardrails I actually stick to.
The result
I ship more in the same hours.
I spend less on AI reruns.
And I end the day feeling like I built something real instead of “being busy.”
If you’re building with Claude/Cursor/Codex all day, try treating money + attention as one budget.
That framing changed everything for me.
Top comments (0)