I thought AI costs and doomscrolling were separate problems — they were the same leak

For months I treated two pains like two separate issues:

• my AI coding bill kept climbing
• my focus kept collapsing into feed refreshes

I kept trying to solve them separately. Bigger prompts for the first. More willpower for the second.

Neither worked.

Then I tracked one simple thing for 14 days: cost per focused build session.

What I noticed

On focused sessions, I wrote shorter prompts, accepted first good outputs, and shipped.
On distracted sessions, I reopened threads, restarted context, retried prompts, and burned tokens.

Same models. Same app. Wildly different cost.

The small workflow that fixed both

1. 90-minute no-feed block before opening AI tools
2. 4-line brief before every run (goal, scope, constraints, done condition)
3. Session cost check every 20 minutes
4. Hard stop when quality drops (usually after context fatigue)

That one system cut retries and made my prompts cleaner.

The tools I ended up building for myself

• TokenBar (tokenbar.site) — a $5 Mac menu bar tracker so I can see AI token/cost drift while I work
• Monk Mode (mac.monk-mode.lifestyle) — a $15 Mac feed-level blocker so I keep utility surfaces (DMs/search) without infinite feed pull

I didn’t build these as “products first.”
I built them because I was tired of paying distraction tax twice: once in attention, once in tokens.

If you’re solo and shipping at night

Don’t ask “which model is cheapest?” first.
Ask: “What part of my workflow creates retries?”

That’s where most of the money leaks.