What is an AI ops retainer?
AI ops retainer definition
An AI ops retainer is a monthly engagement for an AI system that is already in production. Each month the engineer on retainer monitors the system, runs its evaluation harness against the month’s real traffic, updates the models and the integrations it depends on, and ships the next item on an agreed build queue. It is the Operate move of an engagement: what follows once a system is live and the question changes from whether it works to whether it still works, and what it should do next.
An AI system does not stay fixed once shipped. Models are retired and replaced, their behaviour shifts between versions, the APIs around them change, and the data they read drifts as the business does. A system left alone degrades quietly, because a wrong answer looks like any other answer. The retainer exists so that someone reads the evaluation results every month and is paid to act on them.
Why it is not office hours
Office hours sell availability: a number of hours a month, spent on whatever comes up. A retainer in this sense sells a queue. The work for the month is agreed in advance, ranked and shipped, and monitoring and evaluation are standing items that never leave the list. If nothing comes up, the month still produces the next item on the queue. If something breaks, it goes to the top, and the queue moves down.
What a month contains
A review of the system's traces and alerts, and of anything that paged. The evaluation harness run against the month's cases, with any regression fixed before new work starts. Model and dependency updates, each gated by that harness. The next build item, shipped with tests and documentation. And a short written note of what changed, what the numbers say and what is proposed for next month. It is scoped and priced per contract.