DEV Community

Henry Godnick
Henry Godnick

Posted on

The weirdest part of my AI bill was the silence before the spike

A few months ago I started noticing a pattern I didn't like: my AI spend never looked scary in the moment. It looked scary later, when the invoice landed and I had to reverse-engineer what I had done to myself.

That's the part nobody talks about. The problem is not just that AI is expensive. It's that the cost shows up disconnected from the work. You can be heads-down shipping, feeling productive, and still accidentally build yourself a bill that makes no sense.

I hit that feeling hard while building TokenBar, my little menu bar app for tracking LLM usage in real time. I didn't build it because I love dashboards. Honestly, dashboards are where good intentions go to die. I built it because I wanted the number in the place where I actually work: next to the thing I'm using all day.

That changed how I write code.

When token usage is invisible, I keep going. I paste more context. I ask the model to "just refine this one more time." I leave a huge thread open because it feels cheaper than starting over. None of those choices feel expensive in the moment. Together, they absolutely are.

Once I could see usage live, the whole loop got tighter. I started noticing which prompts were bloated, which sessions were eating way more than they should, and which models were overkill for a simple task. The bill stopped feeling like a mystery tax and started feeling like something I could actually manage.

That's why I like the menu bar. It's boring in the best way. It stays out of the way until I need it, and then it tells me the truth.

I think a lot of solo devs are going to run into this same problem over the next year. As more of our stack gets glued together with LLMs, the spend is turning into a real operating cost. Not a maybe. Not a future issue. A line item.

And once that happens, token usage starts behaving a lot like RAM usage. You don't wait until your app crashes to care about memory. You watch it live because the feedback loop matters. AI spend is getting there too.

I made TokenBar because I wanted that feedback loop to be impossible to ignore.

If you're in the same boat, I built it for exactly that reason: https://tokenbar.site

The goal isn't to make you afraid to use AI. It's to make the cost visible enough that you can keep using it without getting blindsided later.

Top comments (0)