A few months into building with Claude, I realized I had a dumb habit: I only checked AI spend after the damage was done.
That usually meant one of two things. Either I had a small bill and felt smug for five minutes, or I had a bigger bill and spent the next ten minutes trying to remember which task had eaten all the tokens.
The worst part was how invisible it all felt while I was working. I could be in the middle of a real coding session, watching a model rewrite the same file three times, and nothing on screen told me I was still paying for the loop.
That is why I built TokenBar. I wanted the number in my face while I worked, not buried in a dashboard I would check later when it was already too late to matter. tokenbar.site
The idea is simple: if AI is part of your workflow, the cost should feel as live as the work itself. Not a spreadsheet problem. Not an end-of-month cleanup task. A live signal, right in the menu bar.
Once I started watching usage in real time, I changed how I worked almost immediately. I stopped letting giant context windows run on autopilot. I noticed when a task was looping. I became a lot less casual about “just one more prompt.”
That is the weird thing about AI spend. It does not usually explode all at once. It creeps.
And if you are a solo dev, “creeps” is another word for money quietly disappearing while you are trying to ship.
I still use AI every day. I just refuse to use it blind.
Top comments (0)