AI Ops
On-call for what the model does, not just whether the server is up.
For teams that already have models in production and no one on call for their behavior. This seat runs versions, cost, eval regression, and incidents when a model update changes what the product does.
The bill doubled and nobody can tell me which agent did it.
What AI Ops owns
Versions
Prompt, tool, and model versions, with regression evals in the release path.
Cost and access
Cost ceilings, loop detection, and who may call what.
Incidents
Response when quality, leakage, or spend moves.
This is a person on a fixed milestone. TokenGuard and Enterprise Agent Governance are the products this seat assumes. The products are tools. The seat is the person on call. IT monitoring, the kind that watches servers, is a different hire.
Where this seat sits
AI Pivot Squad
After the squad ships. This seat runs what the model does.
Business Process AI
After a workflow agent is in production.
Harness Engineer
The harness is what this seat watches. Evals and traces come before on-call.
Common questions
No. IT monitoring watches servers. This seat owns what the model does: versions, cost, eval regression, and the incident when an update changes the product.
No. Those products enforce budgets, catch runaway loops, and keep a registry and a trace. Someone still has to be on call when the number moves. That someone is this seat.
No. Before launch, the Harness Engineer builds the evals and the traces. This seat starts once the model is in production and behavior can move without a deploy of your application code.
Put someone on call for the model.
A 30-minute scoping call is enough to tell whether this seat is the entry, or whether the system is not in production yet.