Your AI Agents Need A Chief Operating Officer, Too
As businesses increasingly adopt agentic AI, effective management requires treating digital agents like human employees by establishing clear rules, variable autonomy levels, and designated human owners.
By Rick Folgmann
A California landlord is legally required to return a security deposit or deliver an itemized list of deductions within 21 days after a resident vacates. Fail to meet this timeline in bad faith, and the penalties can scale up to twice the amount of the deposit. That legal deadline makes no distinction about whether the notification originated from a human worker or a software program. Accountability applies regardless.
This reality is a frequent focus of mine. Within my organization, Get Covered, our operations currently deploy 16 distinct AI agents overseeing upwards of 3 million rental housing units. During the development phase, technical hurdles proved to be the most manageable part of the process. Determining reporting structures for each individual agent presented a far greater challenge.
A similar dilemma awaits most enterprises, yet few preparation efforts are underway. Deloitte’s 2026 State of AI report indicates that 23% of businesses currently utilize agentic AI at least moderately, with 74% anticipating adoption within a two-year window.
Meanwhile, Gartner has projected that more than 40% of agentic AI initiatives will face cancellation by the close of 2027.
Such statistics strongly support managing digital agents with the same framework applied to personnel in regulated positions: documented policies, defined boundaries, direct management, and routine performance evaluations.
The following practices have proven effective for our operations and serve as my primary recommendations for decision-makers managing scale.
Start with work that already has a rulebook
Optimal initial assignments for AI agents involve workflows where established guidelines already exist outside human memory, such as statutory requirements, contract stipulations, or vendor agreement insurance mandates.
Consider vendor certificates of insurance. Across the property sector, management teams dedicate roughly three hours per property weekly tracking down these records via email. While repetitive, the task is fully defined.
Assignments of this nature provide leadership with two distinct benefits. First, the agent has objective criteria to measure its output against, and second, internal teams can audit outcomes without getting bogged down in subjective judgment calls.
Recent State of AI research from McKinsey reveals that the narrow segment of companies achieving substantial financial returns from AI—roughly 6% of surveyed participants—were nearly three times more likely to have fundamentally restructured their workflows.
Based on my professional background, this restructuring begins by documenting the rule explicitly. If a newly hired employee could not comprehend your workflow explanation on day one, an agent will not grasp it either.
Set autonomy by workflow
Technology leaders frequently ask whether a specific AI system possesses autonomy, as though autonomy were an unchanging product feature. In practical application, a single agent should operate under varying degrees of freedom depending on the specific task.
At Get Covered, our configurations vary by agent function. For example, automated reminders sent to vendors whose insurance certificates expire the following month are cleared to run independently. Conversely, automated replies to residents submitting requests for emotional support animals are never permitted.
An effective exercise for leadership groups involves cataloging every workflow an agent touches and evaluating each with a single question: If this specific action contains an error, how long until discovery, and what will the resulting financial or operational cost be? Rapid detection combined with low risk grants an agent broader operational freedom.
Give every agent an owner with a name
Earlier in my professional journey, I learned that every automated credit determination ultimately required an explanation for customers, regulators, or both. Stating that an algorithm made the decision was never considered an acceptable justification.
That principle remains true today, with the added context that modern agents execute actions more frequently and span more systems than historical underwriting models ever did.
Consequently, every agent at our company maintains a designated human owner who holds ultimate responsibility, mirroring how teams or financial budgets are assigned. This individual oversees the error log, determines when to expand or restrict operational boundaries, and answers for performance outcomes.
Establishing clear ownership also prevents a more subtle internal complication. When separate corporate divisions deploy agents independently without coordination, overlapping conflicts often go completely unnoticed.
Where the constraint actually sits
In the contemporary marketplace, continuous technological advancement of models remains certain. However, organizational clarity regarding decision-making authority will not spontaneously improve. Gartner’s cancellation projections highlight this reality: the majority of failing agent projects will falter because organizations failed to define underlying value or manage risk effectively—both of which represent core management responsibilities.
For executives looking toward 2027, my recommendation is to approach each AI agent similarly to a newly hired employee filling a regulated position. Document the operational rules, establish strict boundaries, and appoint a manager responsible for reviewing outputs and discussing performance metrics.
Organizations adopting this methodology will successfully expand agent responsibilities over time backed by verifiable justification. Businesses choosing to bypass these steps will spend 2027 justifying why their software executed unauthorized actions.
Frequently Asked Questions
01What is agentic AI?
Agentic AI refers to artificial intelligence systems designed to act autonomously toward specific goals, executing multi-step workflows, making decisions, and interacting with various software systems with minimal human intervention.
02Why are AI agent projects being canceled?
According to industry forecasts like Gartner’s, many projects fail or face cancellation due to a lack of defined business value, unmanaged operational risks, and absent management frameworks.
03How can companies safely scale AI agents?
Organizations can scale AI safely by starting with rule-based workflows, establishing variable autonomy levels depending on the task, and assigning a specific human owner to monitor and be accountable for each agent’s performance.
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
