I cut my AI coding costs and doomscrolling in 7 days (small habits, big difference)
Published: (March 13, 2026 at 04:41 AM EDT)
2 min read
Source: Dev.to
Source: Dev.to
The Problem
- My AI coding costs were random and stressful.
- As a solo Mac builder, both cost unpredictability and distraction hit hard.
What Worked in 7 Days
1. Make Cost Visible
I built TokenBar – a tiny Mac app that shows live token usage and spend while I code.
TokenBar – tokenbar.site
Big difference:
- Catch runaway sessions faster.
- Tighten prompts earlier.
- Stop “one more retry” loops sooner.
2. Block Distracting Feeds
My worst cost spikes happened after social‑feed breaks, so I blocked feeds during deep‑work blocks.
I created Monk Mode, a Mac app that blocks distracting feeds at the feed level.
Monk Mode – mac.monk-mode.lifestyle
Result: Cleaner, shorter coding sessions.
The Daily Loop
| Phase | Action |
|---|---|
| Before session | • Pick one task • Define a “done” condition |
| During session | • Monitor token burn live (TokenBar) • No feed apps (Monk Mode) |
| After session | • Log what caused the biggest spend • Fix that in the next prompt pattern |
Outcomes
- Fewer zombie coding sessions.
- More predictable daily spend.
- Less context thrash.
- More shipped work.
No magic model switch—just visibility and environment control.
Takeaways
- If your AI bill feels random, don’t try to optimize everything at once.
- Make spend visible while you work.
- Remove feed triggers during build blocks.
That combo changed more for me than any prompt template ever did.
If you’re also building with Claude, Cursor, or Codex and want the exact daily checklist I use to keep both cost and attention under control, let me know.