A few months ago I did the dumb thing every solo dev does: I let the AI bill become “future me”’s problem.
It was small at first. A few dollars here, a little more there. Then I started building faster, using bigger models, keeping more context open, and asking for “just one more change” over and over. By the end of the month the number wasn’t embarrassing because it was huge. It was embarrassing because I couldn’t explain it.
That was the moment it stopped feeling like a bill and started feeling like a leak.
I don’t have a team to absorb that stuff. If I spend $80 on Claude, that’s not some line item buried in company infrastructure. That’s me. That’s pizza for the week. That’s another month of buying ads. That’s the difference between “this side project is fun” and “this side project is quietly draining me.”
So I built a tiny menu bar app called TokenBar for the one place I actually live while working: the top right of my Mac.
Not a dashboard. Not a spreadsheet. Not another tab I’d ignore.
Just a live token counter sitting there while I code, so I can see what I’m spending before I accidentally go full goblin mode on a prompt and burn through half my budget.
That changed the way I work.
I started noticing patterns:
- bigger context usually meant bigger bills
- “quick” AI edits were rarely quick
- I was using the model more like a collaborator than a calculator, which is fine, but expensive
The weird part is how much calmer it feels once the number is visible. It’s the same reason people keep an eye on RAM or battery. You don’t stare at it for fun. You stare at it because surprise costs are stupid.
That’s really why I built TokenBar: token usage is the new RAM usage. If you’re building with AI every day, you should be able to see it in real time.
If that’s your life too, TokenBar is at tokenbar.site.
I’m still obsessed with making AI cheaper to use, not just easier to use. Because “fast” is great until the invoice shows up.
Top comments (0)