I used to think I had two separate problems:
- My AI coding bill kept climbing.
- My focus kept getting wrecked by social feeds.
Turns out it was the same leak: constant context switching.
When I’d bounce between Claude/Cursor/Codex and random feeds every few minutes, I paid twice:
- in token waste (rehashing context)
- in attention reset tax
Here’s the practical system that finally worked for me.
1) Start each AI session with a 60-second scope note
Before I run anything, I write:
- goal
- constraints
- done condition
This alone reduced wandering prompts and “one more try” loops.
2) Use cheap model passes before expensive runs
I now do:
- draft/structure with lower-cost mode
- escalation only for genuinely hard reasoning
My quality didn’t drop. My bill did.
3) Track spend in real time (not at month-end)
The biggest unlock was seeing cost while I work.
I built/started using TokenBar (a $5 Mac menu bar app) to keep token usage visible as I code. When cost is visible, I naturally tighten prompts and avoid wasteful retries.
4) Set a “max retries before rewrite” rule
If I hit 2 failed iterations, I rewrite prompt/context instead of brute forcing a 3rd/4th expensive run.
5) Kill feed-level distractions, not just app-level
Blocking an app is weak when the whole issue is specific feeds.
I use Monk Mode (a $15 Mac app) to block distracting feeds directly so I can still use needed tools without falling into scroll loops.
6) Earn feed time with shipped progress
My rule: no feed opens until one meaningful commit ships.
This sounds strict, but it made work feel cleaner and less guilty.
7) Batch “research scrolling” into a fixed window
If I need X/Reddit/YouTube research, I do it in one short intentional block, then back to execution.
No more all-day half-research, half-distracted state.
8) Run a nightly 3-line review
Every night I log:
- token spend today
- shipped output today
- biggest distraction trigger
After a week, patterns become obvious.
If you’re a solo builder, this combo matters more than any “perfect stack” debate.
You don’t need superhuman discipline.
You need:
- visible cost feedback
- feed friction
- simple operating rules
That’s what finally stopped the bleed for me.
Top comments (0)