A property insurance technology executive is now operating 16 AI agents across more than 3 million rental units, raising questions about accountability when software replaces human staff in regulated industries. Rick Folgmann, chief operating officer at Get Covered, wrote in a Crunchbase News column published this week that the technical challenges of building these agents proved easier than determining who manages them. The essay argues that companies need to treat AI agents like employees in regulated roles, complete with defined rules, assigned managers, and regular performance reviews.
Current adoption of agentic AI stands at 23% of companies using it at least moderately today, according to Deloitte's 2026 State of AI survey cited in the column, while 74% anticipate deploying it within two years. Gartner has forecast that more than 40% of agentic AI projects will be canceled by the end of 2027. Separately, McKinsey research found that roughly 6% of companies are seeing significant bottom-line returns from AI, and those firms were nearly three times as likely to have fundamentally redesigned their workflows. In the property management sector specifically, industry teams spend approximately three hours per property each week pursuing vendor certificates of insurance through email correspondence.
Folgmann recommends starting with tasks that already have external rulebooks—statutes, contract clauses, or insurance requirements in vendor agreements—where agents can check their output against objective standards. Each agent at Get Covered operates with different levels of freedom depending on the workflow it touches: automated reminders to vendors whose certificates expire next month can go out without human review, while responses to residents requesting emotional support animals never do. Every agent has a named person ultimately responsible for reviewing its error log, adjusting its autonomy, and answering for its actions, the same way a team or budget has an owner.
The column frames management clarity as the binding constraint on AI deployment, not model capability. Folgmann argues that most agent projects that fail will do so because nobody defined the value or controlled the risk, framing both as management problems rather than technical ones. Leadership teams should list every workflow an agent will touch and ask how long it would take to detect a wrong action and what it would cost by then—quick detection and low cost justify more autonomy. The example he returns to is California's security deposit law: landlords have 21 days after a resident moves out to return the deposit or send an itemized deduction list, with penalties reaching twice the deposit for bad-faith violations. That deadline doesn't distinguish between failures by people or software.
For decision-makers heading into 2027, the recommended approach is to write down each agent's rules, limit its scope, and assign a manager who checks its work and evaluates its performance. Companies that follow this structure will be able to expand agent responsibility over time because they'll have evidence to justify it. Those that skip the organizational infrastructure will spend 2027 explaining why their software took actions nobody approved. The discipline required to govern autonomous systems effectively will separate operators who can scale AI from those forced to pull back—a distinction that has less to do with technical sophistication than with the clarity of internal accountability structures.