The token spike that told me I was avoiding a decision

I built TokenBar after noticing a weird pattern in my own AI workflow.

Some of my most expensive sessions were not the ones using the biggest model.

They were the ones where I had not made up my mind.

When I am clear, I ask for one thing, evaluate it, and move on.

When I am not clear, I keep the model open like a second brain. I paste more notes. I paste logs I barely read. I ask for three versions, then four more. I compare outputs instead of deciding.

That is when token usage jumps.

The spike is not really about model pricing.

It is often a decision tax.

As a solo founder, that was useful to admit. A lot of AI cost talk focuses on prompts, providers, or context windows. Those matter. But a surprising amount of waste comes from using AI to delay a call that should have been made by the human using it.

Building TokenBar pushed me toward a simpler habit:

• if the token count is climbing fast, I ask what decision I am avoiding
• if I am pasting too much raw context, I stop and summarize it first
• if I need five more generations, the task is probably still fuzzy

That is the real reason I wanted live token visibility in the menu bar.

Not for a dashboard later.

For the uncomfortable moment during the session when I can catch myself turning uncertainty into usage.

That changed how I work more than any after the fact cost report has.

TokenBar is here if you want to try it: https://tokenbar.site/