I cut my AI spend and doomscrolling in one week (solo dev playbook)

Last month I had two ugly habits as a solo builder:

1. I was burning tokens like crazy while vibe-coding.
2. Every time I got stuck, I opened social feeds "for 2 minutes" and lost 45.

If you’re shipping with Claude/Cursor/Codex every day, you probably know this combo.

Here’s the exact playbook I used to fix both in 7 days.

1) I stopped flying blind on AI cost

My biggest mistake was thinking I had a model problem.

I didn’t.
I had a visibility problem.

I kept asking:

• “Why is my bill high?”
• “Was it this prompt or that prompt?”
• “Did I really use that many tokens?”

Without live feedback, every session felt cheap… until invoice day.

What I changed

I started running TokenBar in my menu bar while coding.

It’s a $5 Mac app that shows token/cost usage in real time across my AI sessions.

That one tiny UI change did 3 things:

• I caught waste immediately (looped prompts, retry spam, giant context dumps).
• I switched to smaller prompts faster.
• I treated each run like a budgeted decision, not a casino spin.

Result: lower token burn, same output quality.

2) I killed the feed-level distraction trigger

I used to block whole apps, but that never held up.

I still "needed" X/YouTube/Reddit for work, then got trapped in algorithmic sludge.

So I changed from app-level blocking to feed-level blocking.

What I changed

I started using Monk Mode (my $15 Mac app) to block infinite feeds while keeping useful surface areas.

Examples:

• Keep X compose/search, block home feed.
• Keep YouTube direct video pages, block recommendations.
• Keep Reddit direct thread access, block front page.

That removed the doomscroll entry point without nuking the whole tool.

3) I paired both systems into one daily loop

This was the real unlock:

• TokenBar keeps me honest about AI spend while building.
• Monk Mode protects deep work from feed hijacks.

When cost visibility + attention protection run together, momentum compounds.

My 7-day protocol (copy this)

1. Start day: set one shipping target.
2. Turn on feed-level blocks before coding.
3. Keep real-time token spend visible the whole session.
4. If cost spikes, tighten prompts before continuing.
5. If attention drifts, don’t rely on willpower—let the block hold the line.
6. End day: note what shipped + where time/cost leaked.

What changed for me

After 7 days:

• I shipped more actual features.
• I spent less on AI calls.
• I felt less mentally fried at night.

No revolutionary framework.
Just two constraints:

• Know the cost now (not later).
• Remove feed traps at the source.

If you’re building in public and your velocity feels random, try fixing token visibility and attention hygiene first. Everything else gets easier.