Last week I had a pattern I think a lot of solo devs know too well:
- Spend all day in Claude/Codex/Cursor
- Feel productive
- Check usage/billing later and get punched in the face
- Get stressed, open social apps “for 5 minutes”
- Lose another hour doomscrolling
I finally treated this as one problem, not two separate problems.
The core realization
My money leak and my focus leak were connected.
When I got distracted, my prompts got worse.
Worse prompts = more retries.
More retries = more token burn.
So I ran a 7-day experiment with two tiny constraints:
- Always see token spend while I work
- Block feed-style distraction loops during deep work
My setup (Mac)
1) Real-time token awareness
I kept TokenBar running in my menu bar so I could see cost accumulation while coding.
TokenBar is basically a $5 “speedometer” for AI usage on macOS.
Seeing spend live changed behavior instantly:
- I chunked tasks smaller
- I stopped over-context dumping
- I stopped running expensive “just in case” prompts
2) Feed-level distraction blocking
I used Monk Mode during work blocks to block the infinite-feed parts of sites/apps that usually steal my attention.
Monk Mode is $15 and works at feed level on Mac, which mattered for me because I still needed selective access (messages/docs) without falling into scroll loops.
The 7-day playbook
Day 1–2: Baseline + friction
- Keep TokenBar visible all day
- Turn on Monk Mode for 2 focused blocks/day
- Don’t optimize yet, just observe patterns
Day 3–4: Prompt hygiene
- One objective per prompt
- Hard stop conditions (“if no progress after X, stop and summarize”)
- Smaller context windows unless truly needed
Day 5: Cost caps
- Set a daily “acceptable burn” number
- If I hit it, I switched to cleanup/testing/docs instead of brute-force prompting
Day 6: Protect prime hours
- First 2–3 hours after starting work = Monk Mode always on
- No social feeds until key dev milestone shipped
Day 7: Review
- What burned the most tokens?
- What time windows had worst attention drift?
- Which prompts got repeat retries?
What changed
No miracle, just compounding:
- Fewer chaotic AI sessions
- Less random context thrashing
- Less “I was coding but somehow not shipping” feeling
- Better consistency in actual output
Most importantly, I stopped feeling like my stack was fighting me.
If you’re solo and building
You don’t need a perfect system.
You need a visible feedback loop for both:
- AI spend (TokenBar)
- Attention spend (Monk Mode)
Treat both as one operating metric and your week gets cleaner fast.
If you’ve found other workflows for controlling token costs + attention drift, I’d love to hear them.
Top comments (0)