If you don't trust it to work autonomously, that's a trust problem, not a data problem

Eric MiaoEric MiaoChief Strategy Officer, Attentive

Seventy percent of enterprise marketing, data and AI leaders say their organizations already use custom AI agents in production, according to The Agentic Divide, a survey of 225 senior executives by agentic marketing company Kana. They don't agree on who should own those agents. Forty percent of respondents named the chief AI officer, while marketing executives leaned toward their own function or shared ownership. Agentic AI governance is meant to resolve that disagreement.

What is agentic AI governance?

Agentic AI governance is the set of policies, controls and monitoring that decides what autonomous AI agents can do, which systems and data they can reach, and how their actions are supervised and audited as they happen. That's the definition Attentive uses in its guide. (The CMO Wire is powered by Attentive.)

The term matters now because the tools have changed. Earlier marketing AI recommended an audience or a send time and waited for a person to approve it. Agents act. A campaign agent can build the audience, write the copy, pick the send time and adjust the budget before anyone reviews the result.

How is it different from traditional AI governance?

Traditional AI governance checks a model's outputs for accuracy and bias, with a person approving each decision. Agentic governance has to control actions at the moment they happen. Attentive describes the shift as moving from a human "in the loop," who signs off on each step, to a human "on the loop," who sets the rules, watches the system and steps in when something drifts.

That changes what gets audited. Instead of reviewing a model's predictions after the fact, teams need a record of what each agent did, why it did it and what data it used. Compliance moves from documentation to rules the system enforces in real time, such as quiet hours and frequency caps on text messages.

Eric Miao, then Attentive's chief strategy officer and now its CEO, warned in a September CMO Wire piece that agent failures are harder to see than outages. "A degraded AI system can still be operating while making worse decisions, personalizing content in bizarre ways, or reacting too slowly to matter," he said. "The platform looks fine by every traditional metric."

Who should own AI agents in marketing?

There's no settled answer, which is why governance has to name one. Attentive's governance framework recommends that every agent have its own identity, its own scoped permissions and a single named human owner responsible for its setup, behavior and results.

Ownership also depends on risk. The framework sorts agents into tiers. Read-only reporting agents sit at the bottom. Agents that send customer-facing campaigns sit higher and need stricter permissions, approval steps and closer monitoring. Agents that change customer data or commit spending above a set limit sit at the top.

CMO Wire's recent reporting on how marketing organizations are rethinking AI agent ownership found leaders sorting work into what they hand to agents, what they keep and which function answers for the output.

What should stay with people?

Approval of high-stakes actions, brand standards and the decision about where agents run at all. Attentive's framework calls for human approval before actions that can't be undone or that carry legal or financial exposure.

Miao has argued that the first tasks to hand off are the ones marketers least enjoy. Describing teams that use AI to automate weekly reporting, he told CMO Wire in August, "In that job, zero leverage came from running reporting." Reporting is low risk, so it's also where teams build confidence in agents before giving them customer-facing work.

He also pointed to approval chains as a bottleneck once agents are in place. "If the CEO looks at all of the emails and texts and has to approve them, that's the kind of thing you'll want to change," he said. Governance is what lets a brand remove that kind of review without losing control.

Brand is the area most CMOs keep closest. Miao noted the tension directly. "I don't think Hermès wants to have one-to-one personalization for everything," he said. Liz Villani, founder of Be Yourself at Work, has argued in CMO Wire that brands pay a "verification tax" when AI scales without an authentic human signal behind it.

For Alicia Littleton, director of e-commerce at Gerber Childrenswear, product safety determines how far AI agents can go. "You don't want to accidentally put out there that something is safe for a baby when it's not," she told CMO Wire. "Or little things like an AI suggesting a parent put a blanket in a crib, which is dangerous."

How do teams avoid governance that slows everything down?

By starting with scoped permissions rather than blanket restrictions. Attentive's framework warns that overly tight rules push teams toward one-off exceptions and "shadow agents" that run outside governance entirely. Clear audit trails and well-defined scopes give more control than a long list of prohibitions.

The practical test, in Miao's words from September, is how a vendor answers questions about its safeguards. "If the answer is simply 'trust the model,' keep asking questions," he said. Attentive's guide to implementing agentic AI governance covers how to evaluate a vendor's controls, with marketing examples and common mistakes.

FAQ

What is agentic AI governance? It is the set of policies, controls and monitoring that defines what autonomous AI agents can do, what systems and data they can access, and how their actions are supervised and audited in real time.

How is agentic AI governance different from AI governance? Traditional AI governance reviews model outputs for accuracy and bias, with a person approving decisions. Agentic AI governance controls actions as agents take them, using permissions, real-time monitoring, approval steps for high-risk actions and a record of what each agent did and why.

What does "human on the loop" mean? A human in the loop approves each decision before it happens. A human on the loop sets the rules, monitors the agent's activity and intervenes when something goes wrong, which lets agents act without waiting for approval on routine tasks.

Who should own AI agents in a marketing organization? There is no consensus. In Kana's 2026 survey of 225 leaders, 40% named the chief AI officer, while marketing executives favored their own function or shared ownership. Governance frameworks generally assign each agent a single named human owner, whatever the reporting line.

Which marketing tasks should be handed to AI agents first? Low-risk, repetitive work such as reporting is the usual starting point. Customer-facing tasks like sending campaigns carry more risk and typically need tighter permissions, frequency limits and human approval for high-stakes actions.