I cut my AI coding costs and doomscrolling in the same week (here’s the exact system)

I had a weird month as a solo macOS dev.

On one side: my AI token usage kept climbing.
On the other: my focus kept dropping because every "quick check" of X/YouTube turned into 20 minutes.

I thought these were two separate problems.
They weren’t.

They were both feedback-loop problems.

The old loop (burn money, burn attention)

My day looked like this:

1. Start coding task
2. Open Claude/Cursor/Codex
3. Run big prompt
4. Wait, context-switch to feeds
5. Come back, run another giant prompt
6. Repeat until "workday" is over

Result:

• token spend spiked
• progress felt fake
• mental energy was gone by 3pm

The new loop (ship faster with limits)

I switched to a simple 2-tool system on Mac:

• TokenBar in my menu bar to see token/cost usage in real time
• Monk Mode to block feed-level distractions while I’m in a coding block

I’m not saying "install tools and life is fixed."
The win came from changing behavior because the feedback became immediate.

What changed in practice

1) Cost became visible before damage

With TokenBar running, I could see when a session was getting expensive while it was happening.

So instead of doing 8 speculative follow-ups, I started doing this:

• stop at prompt 2 or 3
• rewrite spec tighter
• restart with smaller context

That alone cut waste hard.

2) Doomscrolling got interrupted at the feed level

I used Monk Mode to block the exact feeds that were killing flow.

Not all websites. Not my entire browser. Just the infinite-scroll surfaces.

That mattered because I still needed docs, GitHub, Stack Overflow, etc.
I just didn’t need algorithmic rabbit holes during deep work.

3) I started measuring sessions, not vibes

Instead of "I worked all day," I now track:

• feature shipped
• token spend for that feature
• distraction incidents during the block

When you track those 3, your process self-corrects fast.

My current workflow (steal this)

Morning setup (5 minutes)

• define 1 shippable target for the day
• open TokenBar
• enable Monk Mode for feeds
• set 90-minute build block

During build block

• max 3 prompt attempts before rewriting spec
• if cost spikes, reduce context instead of brute-forcing
• no feed access until block ends

End of block

• log what shipped
• log rough token cost
• if no ship, inspect where attention leaked

Why this worked for me

Most productivity advice ignores one thing:
attention leaks and token leaks amplify each other.

When focus drops, prompt quality drops.
When prompt quality drops, token spend rises.
Then frustration rises, and you escape to feeds.

Breaking either loop helps.
Breaking both loops compounds.

If you’re building with AI every day

Try this for 7 days:

• keep cost visible (I use TokenBar)
• block infinite feeds during deep work (I use Monk Mode)
• score each day on shipped output, not hours online

You might be surprised how quickly both your burn rate and stress level fall.

If you’ve found a better system, I’d love to steal it.