I kept making the same mistake:
- I treated AI cost as a money problem.
- I treated doomscrolling as a discipline problem.
Turns out they were the same problem: attention leaks.
When my attention was fragmented, my prompts got worse, I retried more, context bloated, and token spend jumped.
The pattern I noticed
On focused build days:
- fewer prompts
- cleaner prompts
- less model hopping
- lower cost per shipped feature
On scattered days:
- endless “one more try” loops
- giant context dumps
- random tab switching
- way higher spend with less shipped
The 7-day reset that worked for me
1) Session budget before I start
I set a rough token/cost budget before opening my editor.
2) Smaller briefs, smaller runs
Instead of one mega prompt, I split work into tiny passes.
3) Hard stop on retries
If I hit 3 failed retries, I pause and rewrite the brief.
4) Feed-free build blocks
I block algorithmic feeds during coding windows.
5) Mid-session cost check
One glance halfway through the session changed behavior fast.
6) Model tiering by task
Heavy reasoning only when needed. Everything else goes cheaper.
7) End-of-day 5-minute review
What shipped, what burned budget, what to fix tomorrow.
The tools I ended up building for myself
I couldn’t find exactly what I wanted, so I made two tiny Mac apps:
- TokenBar — live token + estimated API cost in the menu bar ($5): https://tokenbar.site
- Monk Mode — feed-level distraction blocking on Mac ($15): https://mac.monk-mode.lifestyle
Not magic. Just immediate feedback loops.
Result
I shipped more in the same hours, while spending less on AI runs.
The key shift: stop treating money and attention as separate dashboards.
They’re one system.
If you’re building solo, track both.
Top comments (0)