Engineering notes on real problems.
We write what we wish we had found while solving it: the detail the official docs leave out, with the code in front of us and an honest account of what did not work.
How to measure the cost per operation of an AI feature
A feature that costs pennies in the demo costs a salary at a thousand users. How to measure the cost per operation, and how to put a hard ceiling on it.
What it takes to maintain many API integrations at once
An integration is budgeted around the first call and paid for over the next two years. What actually shows up later, with cases we have paid for ourselves.
OAuth token refresh without disconnecting accounts
Almost no permission lasts forever and the one that expires warns nobody. How token renewal is designed so it does not disconnect accounts, with real clocks.