The AI cost fix that actually changed my behavior as a solo founder
I spent a stupid amount of time trying to reduce AI costs the wrong way.
I kept reviewing logs after long work sessions and telling myself I would write tighter prompts next time. It never stuck.
What finally changed my behavior was seeing token usage while I was still working.
Once the number was always visible, I caught three habits fast:
- I kept carrying too much old context into new tasks.
- I switched to bigger models earlier than I needed to.
- I let sessions run long after the useful part was over.
None of that was obvious from a dashboard later. In the moment, it was obvious.
That was the reason I built TokenBar. I wanted a simple macOS menu bar app that showed token usage live without turning cost tracking into another dashboard I had to remember to open.
The surprising part is that the product did not make me more analytical. It made me more disciplined.
Curious if anyone else building with LLMs has found the same thing: visibility changes behavior more than postmortems do.
