There's a weird parallel between two habits I picked up over the past year: tracking my LLM token usage and tracking what I eat. On the surface they have nothing in common. But the psychology behind why both habits actually stick — and why most people fail at both — is identical.
The Invisible Bleed
Here's the thing about both API costs and calories: they're invisible by default. You don't feel a $0.03 API call. You don't feel 200 calories from a handful of trail mix. Both accumulate silently until you get the bill — either from your credit card or your bathroom scale.
I started paying attention to my LLM costs after a month where I burned through $140 on API calls without noticing. I wasn't doing anything wild — just iterating on prompts, testing different models, running agents. Death by a thousand micro-transactions.
Sound familiar? It's the same reason people underestimate their daily calorie intake by 30-40%. The small stuff is invisible until you make it visible.
Making the Invisible Visible
The fix for both problems is the same: real-time feedback loops.
For token tracking, I built TokenBar — a macOS menu bar app that shows your LLM token usage and costs in real time. Not at the end of the month. Not in a dashboard you'll check once. Right there, always visible, updating as you work.
For nutrition, I switched to photo-based tracking with MetricSync. Instead of manually logging every ingredient (which nobody sustains for more than two weeks), you snap a photo of your meal and AI estimates the macros. The friction drops from 3 minutes to 3 seconds.
Both tools work for the same reason: they turn an invisible process into an ambient, low-friction signal.
Why Most Tracking Fails
Traditional tracking — whether it's calorie counting in MyFitnessPal or checking your OpenAI billing page — fails because it requires conscious effort at exactly the moment you're focused on something else. You're coding. You're eating lunch. You're not thinking about metrics.
The habits that stick are the ones that require zero context switching:
- Ambient visibility — the data is just there, no tab to open
- Minimal input — one glance or one photo, not a form to fill out
- Instant feedback — you see the impact now, not at month end
This is why step counters work but gym logs don't. It's why TokenBar works but billing dashboards don't.
The Compound Effect
After three months of real-time token tracking, I cut my monthly LLM spend by 40% without changing my workflow. I just naturally started choosing smaller models for simple tasks and batching my expensive operations. The awareness alone changed my behavior.
Same with nutrition tracking. I didn't go on a diet. I just started seeing the numbers. That was enough.
The Takeaway for Developers
If you're building tools — especially dev tools — think about ambient feedback over active dashboards. The best tracking tool is one your user never has to remember to open.
And if you're spending more than $50/month on LLM APIs without tracking per-request costs, you're almost certainly leaving money on the table. Same principle as checking your calories: what gets measured gets managed.
I'm a solo dev building native macOS and iOS apps. TokenBar tracks your LLM token usage from the menu bar (tokenbar.site). MetricSync is a photo-based nutrition tracker for iPhone (metricsync.download).
Top comments (0)