Keeping an LLM app current: what a 'model string update' really involved
Updating my Japanese flashcard app exposed problems with retries, image handling and cost tracking. Here's what changed and how I checked it.
9 min read
All writing
Topic: Testing
The 6× gap: reconciling an LLM cost log with the bill
How timed-out requests and retries left costs out of my LLM request log, and the changes I made to track them more accurately.
8 min read
Spec-first, test-first: how I build AI systems with AI coding agents
Written plans, regression tests, agent reviews and checks against real APIs: how coding agents fit into my projects, and what I keep for myself.
12 min read