AI Agents Need COO-Style Oversight Before They Scale

Rick Folgmann, COO at Get Covered, argues that scaling AI agents is less about the model and more about management discipline. His central point: agents working in regulated or operationally sensitive roles need written rules, clear responsibilities and a named human manager accountable for their performance.

He frames the risk with a California landlord example, where a missed security-deposit deadline can trigger penalties regardless of whether the mistake came from a person or software. At Get Covered, Folgmann says the company runs 16 AI agents across more than 3 million rental units, and the hardest question became who each agent reports to.

Folgmann recommends beginning with workflows where the rules already exist in statutes, contracts, insurance requirements or vendor agreements. These tasks give teams an objective standard for judging an agent’s output and make audits less dependent on subjective judgment.

One example he cites is vendor certificates of insurance, which property teams often chase by email and which he describes as tedious but clearly specified. His takeaway is practical: if a new employee could not follow the decision process on day one, an AI agent likely will not follow it either.

The article cautions against treating autonomy as a fixed feature of an AI system. Instead, Folgmann says the same agent should have different levels of freedom depending on the risk, detection speed and cost of a mistake in each workflow.

At Get Covered, a vendor reminder about an expiring certificate can go out automatically, while a response to a resident requesting an emotional support animal does not. Leaders should map every workflow an agent touches and ask: if this action is wrong, how long until we know, and what will it cost?

Folgmann argues that “the model did it” is not an acceptable answer when automated decisions need to be explained to customers or regulators. Every agent should have a person responsible for reviewing error logs, adjusting autonomy and answering for results.

This ownership model also helps prevent agents launched by different teams from contradicting one another. The broader implication is clear: as agentic AI adoption grows, the durable constraint may not be model quality but whether companies define value, control risk and assign accountability.