Henry Godnick

Posted on Mar 13

7 tiny changes that cut my AI coding bill and doomscrolling in the same week

#webdev

Last month I realized I had two leaks happening at the same time:

1) my AI coding costs were creeping up every day
2) my attention was getting shredded by social feeds between work blocks

I kept treating them as separate problems. They weren't.

They were both budget problems.

One was money. One was focus.

Here are 7 tiny changes that fixed both.

1) I stopped doing "just one more" AI run

My most expensive days came from late-night, low-clarity prompts.

Now I write a 3-line brief before every Claude/Cursor/Codex session:

goal
constraints
done condition

That single habit cut retries hard.

2) I put live token spend in my menu bar

If you're building with LLMs daily, delayed billing feedback is brutal.

I now keep TokenBar open all day (a $5 Mac menu bar app that tracks token/cost usage in real time).

Seeing spend while I work changed behavior instantly:

fewer giant context dumps
fewer random model switches
better scoped tasks

3) I switched from "big prompts" to "small loops"

My old style: one huge ask and pray.

My new style:

ask for outline
ask for step 1
verify
continue

Costs dropped and quality improved because I caught mistakes early.

4) I blocked feeds only when I needed to ship

I don't believe in permanent digital monk mode.
I believe in tactical blocking.

I use Monk Mode (a $15 Mac app) to block distracting feeds during deep work windows.

The key thing: it's feed-level blocking, so I can still access useful parts of sites without falling into infinite scroll.

5) I track "cost per shipped outcome"

Raw token count is interesting but not enough.

I now ask: "How much did this shipped feature cost me in AI spend?"

This made me optimize for delivered output, not AI activity.

6) I pair spend limits with attention limits

My rule now:

if I hit my AI spend limit for the block, I stop prompting and implement
if I feel scroll impulse, I extend a Monk Mode block for 30 minutes

Money and focus are now one system.

7) I end each day with a 2-minute review

I log:

what I shipped
what I spent
where I got distracted

Then I adjust one thing for tomorrow.

No giant productivity framework. Just tiny iteration.

TL;DR

If your AI bill is rising and your output is flat, don't just optimize prompts.
Optimize your environment.

DEV Community