I cut my AI tool spend and doomscrolling by tracking one number per session

For months I treated these as two separate problems:

1) My AI coding costs kept creeping up
2) My focus was getting wrecked by feed scrolling between tasks

Turns out it was the same leak.

When I was distracted, my prompts got sloppier, retries increased, and each "quick" coding session became expensive.

The one metric that changed everything

I started tracking one number after every build session:

cost per shipped session

Not cost per month. Not cost per model. Cost for a session that actually shipped something useful.

That instantly exposed the difference between:

• focused sessions (cheap + high output)
• scattered sessions (expensive + low output)

My current 20-minute reset routine

Before I start coding:

• define one concrete output for the next block
• set a soft budget for the session
• remove feed triggers during the block

After I finish:

• log spend
• log what shipped
• note if retries came from unclear prompts or context bloat

Boring system, but it works.

Tools I built for this workflow

I built two tiny Mac apps around this exact loop:

• TokenBar ($5): real-time token/cost visibility in the menu bar while I work
  
  https://tokenbar.site

• Monk Mode ($15): feed-level distraction blocking so I can keep useful app features without the infinite scroll trap
  
  https://mac.monk-mode.lifestyle

They’re simple, but together they close the loop:
see spend → protect focus → ship more → spend less.

Biggest lesson

Most of my AI bill wasn’t from "bad model pricing."
It was from broken workflow hygiene.

If your costs feel random, try this for 7 days:
track cost per shipped session and remove feed triggers during build blocks.

You’ll probably find the same thing I did: attention debt becomes token debt.