Service 06
AI Ops & Maintenance
Once it is live: evals, drift, cost, logs — the unromantic part.
- Monitoring
- Cost
- Regressions
Launch is the start of the expensive half.
Models change underneath you. Providers deprecate versions, prices move, users find inputs nobody imagined, and quality drifts in a direction nobody is measuring. A system nobody is watching is a system getting quietly worse.
This is the arrangement that keeps the lights on: the eval suite running on a schedule, cost visible per request and per month, logs you can search when somebody asks what happened, and a person to call.
What we actually do
- The eval suite run on a schedule, with a number that moves when quality does
- Cost per request and per month, tracked and alerted on
- Structured logs and traces, kept long enough to answer a question
- Provider changes absorbed — deprecations, model swaps, fallbacks
- Regressions found by us, before they are found by your users
- A monthly note in plain language: what changed, what it cost, what we’d do next
Who it’s for
Anyone running a system in production that nobody currently has the time to watch.
How it runs
A monthly arrangement, agreed after a project rather than as a condition of one. Cancellable, and never a way of keeping hold of your accounts.
Start with the problem.
Tell us what you’re working on and whether ai ops & maintenance is really what it needs. An engineer reads every enquiry.