For months I thought I had two separate problems:
- my AI coding bill kept swinging wildly week to week
- my focus kept collapsing at night
Turns out it was one system bug.
Every time I context-switched into algorithmic feeds during a build session, I came back and wrote sloppier prompts, reran tasks, and burned extra tokens.
So I started tracking only two metrics every night:
- AI spend per session (not monthly total)
- feed-triggered context switches (how many times I got pulled out of flow)
That simple scoreboard changed everything.
What I changed
1) Session spend visibility in real time
I stopped waiting for end-of-month surprises.
I built TokenBar (https://tokenbar.site) as a tiny Mac menu bar tracker so I can see token/cost impact while I work.
Price: $5 one-time.
The biggest shift: I finally noticed that my expensive sessions weren’t always my longest sessions — they were the sessions with the most retries.
2) Remove feed triggers, not useful app features
Deleting apps never lasted for me.
I built Monk Mode (https://mac.monk-mode.lifestyle) to block algorithmic feed surfaces while keeping useful parts of apps available.
Price: $15 one-time.
This gave me long uninterrupted coding blocks without going full “digital monk.”
3) End-of-night reset (15 minutes)
Before sleep:
- close stale agent loops
- note top 3 tasks for tomorrow
- cap unfinished experiments
- clear feed-trigger tabs
The morning handoff got cleaner, and rerun-spend dropped.
What changed in practice
- fewer “what was I doing again?” moments
- fewer panic retries after outages/context loss
- lower token spend volatility
- more actual shipping, less fake busyness
If you’re a solo dev feeling both token burn and attention burn, try treating them as one pipeline issue, not two separate habits.
Track the money where the work happens, and remove the triggers that break flow.
That’s the whole game.
Top comments (0)