DEV Community

Henry Godnick
Henry Godnick

Posted on

I Was Burning Money on AI and Attention — Here’s the 7-Step System That Fixed Both

If you build with AI every day, two leaks quietly kill momentum:

1) You overspend on tokens without noticing in the moment
2) You lose deep-work time to feed scrolling between prompts

I hit both hard this month.

Here’s the exact system I now use to ship faster with less stress.

1) Track spend while you work, not after

End-of-day invoices are too late. Real-time feedback changes behavior.

I keep a tiny menu bar tracker running so I can see usage as I iterate. If I see a spike, I tighten prompts immediately instead of discovering it tomorrow.

2) Split “explore” and “execute” sessions

Exploration is intentionally messy. Execution should be tight.

  • Explore: broad brainstorming, rough ideas, dead ends allowed
  • Execute: scoped tasks, short prompts, deterministic output

Mixing these is how token bills explode.

3) Use a hard budget per coding block

Before a session, I set a simple cap for that block.

When I approach it, I switch from asking huge open prompts to smaller surgical ones. Same progress, lower burn.

4) Kill the feed during deep work

My biggest cost spikes often happened after “just a quick scroll.”

Context switching made me re-prompt everything, which increased both time and token spend.

So during coding blocks, I block distracting feeds at the source (not entire websites, just the feeds).

5) Keep a running prompt log

When a prompt pattern works, save it. When one wastes tokens, tag it.

A lightweight log quickly becomes your personal playbook and prevents repeating expensive mistakes.

6) Design your environment for low-friction focus

Tiny defaults matter:

  • AI cost meter always visible
  • Feed blockers scheduled for work windows
  • Notes window pinned for task scope

Good defaults beat willpower.

7) Review once per week

Every week, I check:

  • Which sessions had the worst cost-per-useful-output?
  • Which distractions broke flow most often?
  • Which prompt templates gave best outcomes?

Then I update my workflow for next week.


The stack I’m currently using

  • TokenBar ($5): Mac menu bar tracker for live AI token/cost visibility while you code.
  • Monk Mode ($15): Mac distraction blocker with feed-level blocking so you can keep tools open but remove doomscroll traps.

The combo works because it attacks both leaks at once: money leak + attention leak.

If you’re serious about shipping consistently with AI, this pair changed the game for me.

Top comments (0)