- Running agents at scale is a standing operational function, not a project.
- Most operators will not staff it internally, and the managed-services category absorbs the work.
- What an MSP cannot absorb is the authority decisions. Those stay with you.
Most operators will never build the in-house capability to run agentic systems at scale. Not because they lack ambition, but because the capability is a standing function with specialist staffing, and it competes for budget against the business the company is actually in. That gap is what the managed-services category absorbs.
Why running agents is a standing function
An agent in production needs the same care as any other operating system component, plus some that are new. Model behaviour drifts. Upstream systems change shape. Exception rates move when the business does. Permissions need review as scope evolves.
That is monitoring, tuning, incident response and periodic governance review, continuously, per agent. It is not a project that finishes.
Multiply by a growing agent estate and you have a team. Most mid-market operators will not build that team, and should not pretend the requirement disappears as a result.
What the category absorbs, and what it cannot
A managed provider can absorb the operational surface: running the agents, watching the telemetry, handling incidents, applying updates, keeping the audit trail complete, and reporting on it.
What cannot be outsourced is the authority layer. Who delegated this agent's authority, what it may decide alone, and who revokes it are answers about your accountability, and a provider cannot hold them for you.
The failure mode is a relationship where operations are outsourced and authority quietly goes with them. Then nobody in your organisation can say what the agents are permitted to do.
Most operators won't build the in-house capability to run agentic systems at scale. The MSP category absorbs the operating-model work.
Structuring it without hollowing out your team
Keep three things in-house even in a fully managed arrangement: the authority definitions, the exception escalation path, and the measurement you use to judge the provider.
That is a small retained capability — often one person with real authority rather than a team — and it is what separates a managed relationship from a dependency.
Write it into the contract, because the drift happens by default, not by decision.
What to do next
- 01Decide explicitly whether agent operations is a function you will staff. Ambiguity here becomes a dependency.
- 02Retain the authority definitions, escalation path and measurement in-house regardless of who operates.
- 03Name the internal owner of agent authority in the managed-services contract itself.