For months I thought I had two unrelated issues:
1) my AI coding costs were all over the place
2) I kept losing hours to feed-driven context switching
Turns out it was the same bug.
When I got distracted, my prompts got sloppy.
Sloppy prompts meant retries.
Retries meant token burn.
What changed
I started tracking spend per session and focus quality together.
- If a session was expensive, I checked where my attention broke.
- If I had a doomscrolling day, I expected higher AI spend.
- If I stayed in one deep block, cost per shipped task dropped.
That one framing shift changed everything.
My current stack (simple, cheap, repeatable)
1) Token visibility in real time
I keep a live token/cost counter in my menu bar with TokenBar (https://tokenbar.site, $5).
It sounds tiny, but seeing cost while I work prevents the “I’ll clean this up later” lie.
2) Feed-level distraction blocking
I stopped trying to delete apps and just removed the feed surfaces with Monk Mode (https://mac.monk-mode.lifestyle, $15).
I can still use utility parts of apps. I just don’t let infinite feeds hijack a build window.
3) A 2-minute preflight before heavy runs
Before long AI runs, I write:
- goal
- done condition
- max retries
- fallback model
This alone cuts panic loops.
Results from this workflow
- fewer late-night “what did I even ship?” sessions
- fewer expensive retries
- better output quality from shorter, cleaner prompts
- less emotional thrash
No magic. Just better feedback loops.
If you’re a solo builder, don’t separate productivity from cost control.
They’re the same system.
Top comments (0)