# Enterprises Must Appoint Human Overseers For AI Agents, Not Just Deploy Them Autonomously
Rick Folgmann, COO at Get Covered, argues that companies treating AI agents as fire-and-forget tools will stumble once these systems handle regulated workflows or touch customer-facing operations. The solution mirrors corporate governance: assign a human manager as accountable owner, codify operating procedures, and calibrate autonomy based on task risk.
The emerging AI agent market has pushed enterprises toward rapid deployment. Vendors pitch autonomous systems that learn, adapt, and operate without constant supervision. The pitch resonates because autonomous agents promise efficiency gains and reduced headcount costs. But Folgmann's thesis pushes back on deployment theater. Slapping an AI agent on a workflow without governance structures produces liability exposure, data leaks, and regulatory violations.
His framework starts simple. Companies should begin with rule-governed tasks where outcomes map to clear policies. Insurance underwriting, claims processing, and compliance documentation all fit this profile. These domains have established playbooks. An AI agent executing pre-approved rules within defined guardrails reduces operational risk compared to an agent making novel decisions on unstructured problems.
The accountability layer differs from traditional software. A buggy app gets patched. An agent that breaches compliance or mishandles a customer escalates differently. Someone owns that outcome. Folgmann advocates naming a human manager responsible for the agent's performance, similar to how companies appoint a product manager for a business line. That person sets the agent's scope, reviews its decision-making patterns, and recommends scope expansion or rollback.
Risk calibration comes next. A customer service chatbot handling FAQ routing requires looser constraints than a loan underwriting agent. Folgmann suggests auditing each workflow's downside scenario, then setting autonomy ceilings accordingly. A low-risk task might grant the agent 100 percent decision authority. A high-risk task might cap autonomy at 70 percent, requiring human sign-off on edge cases. Workflows involving regulated industries like finance, healthcare, or insurance need tighter controls than internal productivity tools.
Regular audits matter more than most enterprises realize. Folgmann recommends systematic reviews of agent decisions, error rates, and customer complaints. These audits feed into expansion decisions. If an agent operating at 60 percent autonomy maintains a 99.2 percent accuracy rate for six months, the human manager can justify increasing autonomy to 75 percent. That progression prevents catastrophic failure and builds institutional confidence.
The comparison to employee onboarding resonates. New hires start with training, clear job descriptions, and a manager. They earn expanded responsibility through demonstrated competence. AI agents should follow the same progression. Many enterprises skip these steps, treating agents as fully formed performers. That's a recipe for operational chaos.
This perspective reflects a broader shift in enterprise AI strategy. Early-stage agent companies like Anthropic, Retool, and Zapier are embedding human-in-the-loop controls directly into their platforms. Enterprises deploying agents at scale report similar conclusions independently. The winners treat agents as specialized employees, not autonomous robots.
The COO framework also addresses board-level concerns about AI governance. Risk committees and compliance officers now ask pointed questions about agent oversight. Having a named manager and auditable decision trails converts those questions into checkboxes rather than open debates.
Folgmann's recommendations don't eliminate AI agent adoption. They channel it responsibly. Companies that implement this structure early will move faster than competitors scrambling to retrofit governance into already-deployed systems.