As AI applications become more complex, are people really tracking token usage and costs at the workflow level?
It’s easy enough to see the use for individual model calls, but once a feature spans multiple signals, models, tools, retries, and background jobs, I find it much harder to answer questions like these:
-
Which workflow is driving up costs?
-
Where is latency being introduced?
-
Which step failed?
-
How much does a single user action really cost?
Want to know what other people are using today.
<a href