Your AI Agents Also Require A Chief Operating Officer

By Rick Folgmann

California law gives landlords a strict 21-day window after a tenant moves out to either return the security deposit or provide an itemized list of deductions. Acting in bad faith and missing this deadline can result in penalties reaching twice the original deposit amount. The law does not differentiate between notices generated by human hands or software programs; accountability remains the same regardless.

This reality is a frequent point of consideration at Get Covered, where we currently deploy 16 AI agents across a portfolio exceeding 3 million rental units. Technical hurdles proved to be the more manageable aspect of their development. Determining the reporting structure and direct management for each individual agent presented the true challenge.

Organizations everywhere are about to confront this exact dilemma, yet preparation remains scarce. Findings from Deloitte’s 2026 State of AI survey indicate that 23% of businesses currently utilize agentic AI moderately, while 74% anticipate doing so within a two-year timeframe.

Meanwhile, Gartner has forecasted the cancellation of more than 40% of agentic AI initiatives by the close of 2027.

Such statistics underscore the necessity of managing AI agents similarly to staff in heavily regulated positions—complete with documented policies, clear boundaries, designated oversight, and routine evaluations.

Outlined below are the practices that yielded success for our organization, alongside recommendations for leaders aiming to scale effectively.

Start with work that already has a rulebook

Initial agent implementations should target tasks governed by external, established guidelines, such as statutes, contract stipulations, or vendor agreement insurance mandates.

Consider vendor certificates of insurance. Property teams across the sector typically dedicate about three hours weekly per property to pursue these files via email. While the task is repetitive, its parameters are entirely predefined.

Such duties offer leaders two distinct advantages: the agent possesses an objective standard for output verification, and internal teams can audit outcomes without getting bogged down by subjective interpretations.

According to recent State of AI research from McKinsey, the narrow cohort of businesses generating substantial bottom-line gains from AI—roughly 6% of participants—were nearly three times more likely to have completely overhauled their operational workflows.

Based on personal experience, this restructuring begins with documenting the rule itself. If a newly hired employee fails to comprehend a process based on your initial explanation, an AI agent will not succeed either.

Set autonomy by workflow

Technology executives frequently inquire whether a specific AI platform is autonomous, treating the designation as an unchanging product feature. In practice, an individual agent requires varying degrees of independence dictated by the specific task it handles.

At Get Covered, agent configuration involves dynamic permission levels. For example, automated dispatches may notify vendors regarding certificates expiring the following month. Conversely, inquiries from residents seeking accommodations for emotional support animals never bypass manual review.

Leadership groups benefit from evaluating every agent-managed workflow by posing a specific question: Should an error occur during this action, what is the timeline for discovery, and what will be the associated financial impact? Rapid detection combined with minimal risk justifies granting an agent greater independence.

Give every agent an owner with a name

Earlier professional experiences reinforced a fundamental principle: automated credit decisions invariably require justification to customers, regulatory bodies, or both. Claiming that the algorithm executed the action has never served as a valid defense.

This standard persists today, magnified by the fact that modern agents operate with higher frequency and across broader system networks than previous underwriting models ever did.

Consequently, every agent within our firm maintains a designated human owner, mirroring the accountability assigned to standard teams or budgets. This individual reviews error logs, adjusts operational boundaries, and takes full responsibility for performance.

Direct ownership also resolves a subtle complication. When diverse departments deploy agents autonomously without coordination, overlapping contradictions often go unnoticed.

Where the constraint actually sits

Continuous technological advancement in models remains an inevitability of the modern marketplace. However, organizational clarity regarding decision-making authority does not spontaneously evolve. Gartner’s projections regarding project terminations highlight this dynamic: failed agent initiatives typically stem from a failure to define value or mitigate risk—both of which represent core management responsibilities.

For executives navigating toward 2027, the primary recommendation is to approach every AI agent as if hiring a new employee for a regulated role. Establish documented guidelines, restrict operational parameters, and assign a supervisor to audit performance and review metrics.

Adopting this methodology enables enterprises to expand agent autonomy safely over time, backed by empirical justification. Conversely, businesses bypassing these safeguards will spend 2027 justifying unauthorized software actions.

Rick Folgmann is chief operating officer at Get Covered, an insurance technology company that provides cost-effective property coverage with just a few clicks. He has more than 25 years of executive leadership experience in the financial and insurance industries.

Illustration: Dom Guzman