When an AI agent holds a real job, running your demos, onboarding your accounts, someone has to hold the job of managing it, and most orgs discover they never assigned one. Reporting lines for non-human employees sound like a philosophy seminar and are actually an ops question with a concrete answer: an owner, a review cadence, and an escalation contract.
Why "nobody manages it" is the default failure
An unmanaged agent doesn't fail loudly; it plateaus quietly. The agenda drifts out of date as the product and pitch evolve, the question themes nobody reads stop informing anyone, guardrails never widen because nobody's accountable for widening them, and after two quarters the tool "didn't live up to it," a performance review no one was assigned to conduct. The mechanism rewards management exactly the way people do: attention, feedback, and scope granted against evidence. Configure-and-forget is the whole category's most common failure mode, and it's organizational, not technical.
Who should own the agent?
The manager of the function, not the buyer of the tool. The demo agent reports to whoever owns pipeline generation; the onboarding agent to the head of CS or onboarding; the support agent to support leadership; the training agent to enablement or customer education. The rule underneath: the owner is the person whose outcome metric the agent moves, because that alignment is what makes the review honest, the demo agent's owner reads qualified-meeting numbers the way they'd read a rep's. IT and ops are stakeholders (governance, integrations, security posture), not managers; the platform admin is the agent's HR system, not its boss.
What managing one actually involves
A weekly review, thirty minutes, same shape every time: outcomes against the lane's metric, escalation audit (right triggers? full context? anything that should widen or narrow?), and question themes, the free research stream that names next week's fixes to agendas, docs, or product. Quarterly: scope review, what the agent has earned the right to do more of, the way you'd widen any strong performer's lane. And when the team includes several agents sharing account memory, someone owns the team view: the handoffs, the account-level picture, usually whoever owns lifecycle or revenue operations. Write these cadences down when you deploy; the org chart entry is the deployment artifact nobody ships and everybody needs.
What changes for the humans nearby
Job descriptions shift honestly: the onboarding specialist's week moves from delivering sessions to designing them and reviewing the evidence; the SE's calendar purifies toward judgment work; team leads gain a direct report that documents everything and never resents feedback. The teams that thrive treat the agent's arrival like a headcount event, announced, owned, reviewed, rather than a feature flag. (Disclosure: Skippr's agents are built to be managed, outcome metrics per role, documented escalations, reviewable sessions, and the weekly-review habit above is the strongest predictor of compounding value we can point to.)
Questions buyers actually ask
Does the agent need a performance review?
It gets one whether you schedule it or not, at renewal, with worse data. The weekly half-hour is the cheap version.
Who handles the agent's mistakes?
The owner triages like any team member's: fix the grounding or agenda if teachable, narrow scope if structural, escalate to the vendor if platform-level. The audit trail makes attribution unambiguous.
Should agents appear on the org chart?
Practically, yes, an owner, a role, a metric. Whether you draw the box is culture; that someone owns the box is operations.
See what a live agent actually does
The category is easier to watch than to define. Fifteen minutes is enough to see where the mechanism differs from everything it gets confused with.