Most of the work in an AI product isn't the model call. It's the evals, the logging, the orchestration, the budget math, and the decisions about what to build versus what to buy.
These posts are for the teams doing that work — especially small ones without a platform team to hand it to.
Six fields, one JSONL file, first deploy. The instrumentation playbook that answers every question you'll be asked about your agent in month three.
Pipeline, fan-out/fan-in, escalation ladder, verify loop. Each pattern has a signature failure: a place the bill or the latency hides until production.
Thirty real queries with expected answers will save you more money than any discount negotiation. You can't swap models, cut costs, or route traffic without a way to score the output.
The demo is the easy 20%. This is the other 80% — seven checks that decide whether your AI feature survives contact with real users and a real invoice.