For months I thought I had two different problems:
1) my AI coding spend was unpredictable
2) my focus was wrecked by feeds and context switching
Turns out it was the same system failure.
When my attention was fragmented, my prompts got lazy, I retried more, and my token burn spiked. On focused days, outputs were cleaner and I shipped faster with fewer reruns.
What changed for me
1) I added real-time cost visibility
I built TokenBar (https://tokenbar.site), a $5 Mac menu bar tracker that shows token/cost flow while I work.
Not glamorous, but seeing spend live changed behavior immediately.
2) I stopped “all-or-nothing” blocking
I built Monk Mode (https://mac.monk-mode.lifestyle), a $15 Mac app that blocks algorithmic feeds instead of nuking entire apps.
That meant I could keep useful surfaces (messages, uploads, docs) while removing infinite scroll triggers.
3) I started running short session postmortems
At the end of each block I log:
- what I shipped
- what model was used
- where retries happened
- what distracted me
This made costs and attention measurable in one place.
The practical loop
- 90-minute focused block
- feed-level blocking ON
- live cost visible in menu bar
- 5-minute postmortem
Repeat 2–3 times/day.
What happened
- fewer panic reruns
- cleaner prompts
- more predictable AI spend
- more actual shipped work
If you’re solo and your bill feels random, don’t only optimize model choice.
Optimize attention hygiene first.
Top comments (0)